/h/ - Hentai » Thread #7676457

[1279 / 302]

883KiB, 1064x1200, 00238-1203478796thread.png

/hdg/ - Hentai Diffusion General #387

Anonymous Mon 27 Nov 2023 00:44:06 No.7676457 View View Reply Original Report

Quoted By:

Previous Thread: >>7672874

>WEBUI SETUP
Nvidia: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-NVidia-GPUs
AMD: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs
Cloud: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Online-Services
Optimizations: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Optimizations

>RESOURCES
Wiki: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki

LoRAs/models: https://civitai.com/ | https://huggingface.co/ | https://gitgud.io/gayshit/makesomefuckingporn
Training: https://github.com/derrian-distro/LoRA_Easy_Training_Scripts | https://rentry.org/59xed3

Tags: https://danbooru.donmai.us/wiki_pages/tag_groups
ControlNet: https://rentry.org/dummycontrolnet
LamaCleaner: https://huggingface.co/spaces/Sanster/Lama-Cleaner-lama | https://lama-cleaner-docs.vercel.app/install/pip
Animation: https://rentry.org/AnimAnon
Wildcards: https://rentry.org/NAIwildcards

4chanX Catbox / NAI prompt userscript: https://rentry.org/hdgcb
Booru: https://aibooru.online

OP Template/Logos: https://rentry.org/hdgtemp/raw | https://files.catbox.moe/xk99gy.png | https://files.catbox.moe/35bdt5.gif

>Related
>>>/aco/sdg
>>>/g/sdg
>>>/d/ddg
>>>/e/sdg
>>>/vt/vtai
>>>/vg/aids
>>>/b/degen
>>>/trash/sdg

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw,1girl, ai-generated, light (...).png, 687KiB, 1216x832

Anonymous Mon 27 Nov 2023 00:53:36 No.7676473 Report

Quoted By: >>7676477 >>7676482 >>7676483 >>7676558 >>7676860

>The ai-generated tag in NAI perfectly emulates the fucked up AOM anatomy.

I'm probably late to the party, but my sides are in fucking orbit.

Anonymous

View Same Google ImgOps iqdb SauceNAO 92a48aa7-1e70-49d7-87b1-07a3aefc (...).png, 2MiB, 1856x1280

Anonymous Mon 27 Nov 2023 00:54:58 No.7676474 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 00:57:16 No.7676477 Report

Quoted By:

>>7676473
ok thats pretty funny

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, kama (fate), (...).png, 2MiB, 1024x1024

Anonymous Mon 27 Nov 2023 00:59:33 No.7676480 Report

Quoted By: >>7676641

Anonymous

View Same Google ImgOps iqdb SauceNAO ai-generated, solo, 1girl, best (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 01:00:36 No.7676482 Report

Quoted By:

>>7676473
sovl

Anonymous

Anonymous Mon 27 Nov 2023 01:00:41 No.7676483 Report

Quoted By:

>>7676473
incredible lmao

Anonymous

Anonymous Mon 27 Nov 2023 01:01:48 No.7676486 Report

Quoted By: >>7676489 >>7676495

just how much fucking damage did aom2 do to local gen?

Anonymous

Anonymous Mon 27 Nov 2023 01:03:12 No.7676489 Report

Quoted By: >>7676495

>>7676486
that one dude with his shitty lora tutorial poisoned the well

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{{blank eyes, white eyes}}}}, (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 01:03:16 No.7676490 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 1663234777499-1247694729.png, 1MiB, 1152x1408

Anonymous Mon 27 Nov 2023 01:03:19 No.7676491 Report

Quoted By: >>7676523

Choco thread

Anonymous

Anonymous Mon 27 Nov 2023 01:07:56 No.7676495 Report

Quoted By: >>7676498

>>7676486
>>7676489
Not as much as NAI CLIP skip.

Anonymous

Anonymous Mon 27 Nov 2023 01:10:31 No.7676498 Report

Quoted By:

>>7676495
How did that poison anything?

Anonymous

Anonymous Mon 27 Nov 2023 01:12:34 No.7676500 Report

Quoted By: >>7676503 >>7676506 >>7676536 >>7676539 >>7678357 >>7678438

Shillin' my shitty NAI API for A1111 extension again, if anyone wants to pretend they're running local while genning with NAI, give it a try.

https://github.com/Metachs/sdwebui-nai-api

Supports most NAI features, including inpainting, and lets you generate XYZ grids and use wildcards.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00115-1741687393.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 01:17:52 No.7676503 Report

Quoted By:

>>7676500
thanks again for extension anon, i've been genning non-stop since you released it with wildcards IoI

Anonymous

Anonymous Mon 27 Nov 2023 01:18:35 No.7676506 Report

Quoted By:

>>7676500
I'm stuck on a macbook for the time being and I hate the NAI UI, there's so much wasted space
Hopefully I can install the webui on here and get your extension to work

Anonymous

Anonymous Mon 27 Nov 2023 01:20:40 No.7676510 Report

Quoted By:

>>7676463
>I guess there's no point in sticking around now that nai figured things out, I'm moving on into text gen.
same, im hoping that text gen is not the clusterfuck that i saw half a year ago

Anonymous

Anonymous Mon 27 Nov 2023 01:27:16 No.7676515 Report

Quoted By: >>7676519 >>7676527 >>7676528 >>7676530 >>7676541

what if /hdg/ had exclusive access to a GH200 for some reason?
would there be enough dataset autists here to make use of it?

Anonymous

Anonymous Mon 27 Nov 2023 01:31:22 No.7676519 Report

Quoted By: >>7676556

>>7676515
I'd do it FOR FREE
Memes aside we'd need to establish a proper team - the single biggest hurdle when you're on here. I know jack and shit about training a model, I've barely trained a few good enough LoRAs but I've seen enough hentai to know what I'd want in a good dataset. We'd absolutely need to ditch regular tagging though.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, olivia (pokem (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 01:34:13 No.7676523 Report

Quoted By:

>>7676491
I'm ok with this theme

Anonymous

Anonymous Mon 27 Nov 2023 01:38:00 No.7676527 Report

Quoted By:

>>7676515
is that what nai used?

Anonymous

Anonymous Mon 27 Nov 2023 01:38:36 No.7676528 Report

Quoted By:

>>7676515
yeah

Anonymous

Anonymous Mon 27 Nov 2023 01:39:20 No.7676530 Report

Quoted By: >>7676535

>>7676515
dataset yes, but tagging?

Anonymous

Anonymous Mon 27 Nov 2023 01:42:54 No.7676535 Report

Quoted By: >>7676546

>>7676530
dataset isn't a dataset without labels me thinks

Anonymous

Anonymous Mon 27 Nov 2023 01:43:47 No.7676536 Report

Quoted By:

>>7676500
Most useful thing ever. Thank you

Anonymous

Anonymous Mon 27 Nov 2023 01:46:57 No.7676539 Report

Quoted By:

>>7676500
>if anyone wants to pretend they're running local while genning with NAI
is there a way of just integrating nai into workflow?

Anonymous

Anonymous Mon 27 Nov 2023 01:47:04 No.7676541 Report

Quoted By: >>7676559

>>7676515
dataset is only partially the problem.
Tagging is in a weird place right now in terms of what is and isn't best practice and the actual training methodology is nebulous tier at best.
And while I'd want to think it wouldn't be the worst idea to try and 'crowdsource' a curated dataset you'll probably just end up with an issue of a lot of subpar images(due to the erratic nature of what people consider to be good or bad quality) on top of just not actually getting that large of a dataset in the end.

Anonymous

View Same Google ImgOps iqdb SauceNAO K1.png, 388KiB, 512x768

Anonymous Mon 27 Nov 2023 01:48:54 No.7676544 Report

Quoted By: >>7676549 >>7676910

>>7674281
>>7674284
>>7674312
>>7674216
>NAI3 can do Persona
I don't suppose you can help an anon out and make some Kasumi gens? Sorely lacking in art of her.

Anonymous

Anonymous Mon 27 Nov 2023 01:50:27 No.7676546 Report

Quoted By:

>>7676535
correct

Anonymous

View Same Google ImgOps iqdb SauceNAO K2.png, 517KiB, 768x512

Anonymous Mon 27 Nov 2023 01:52:44 No.7676549 Report

Quoted By: >>7676552

>>7676544
I keep trying to make an evening scene under the covers but it always ends up lacking.

Anonymous

View Same Google ImgOps iqdb SauceNAO 558-nsfw,1girl,by Yoji Shinkawa, (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 01:52:52 No.7676551 Report

Quoted By: >>7676557

I love that pistols just work in NAIv3, it can even get some neat poses. Too bad long guns don't work nearly as well.

Anonymous

View Same Google ImgOps iqdb SauceNAO K3.png, 356KiB, 768x448

Anonymous Mon 27 Nov 2023 01:53:54 No.7676552 Report

Quoted By: >>7676554 >>7677253

>>7676549
Sometimes turns out okay with other poses but I'd still love to see how what I do compares with what NAI3 by default.

Anonymous

Anonymous Mon 27 Nov 2023 01:55:14 No.7676553 Report

Quoted By:

nai can't do hanabatake yoshiko

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, yoshizawa kasumi, persona (...).png, 2MiB, 1216x832

Anonymous Mon 27 Nov 2023 01:55:52 No.7676554 Report

Quoted By: >>7676566

>>7676552
Here is something very quick done from a very incomplete prompt. I'll keep playing around for a bit.

Anonymous

Anonymous Mon 27 Nov 2023 01:56:29 No.7676556 Report

Quoted By:

>>7676519
We would need a discord server or another place for this to happen honestly, here we are so much disorganized

Anonymous

Anonymous Mon 27 Nov 2023 01:57:01 No.7676557 Report

Quoted By: >>7676571

>>7676551
long guns are always a crapshoot because of the way people draw them and how badly they're described. pistols are pistols and sometimes minor details get weird but they're on average drawn and tagged the same, but with long guns, anything from a over/under shotgun to an LMG to a sniper rifle and rocket launcher
might be tagged the same and it fucks everything up in the end

Anonymous

Anonymous Mon 27 Nov 2023 01:57:45 No.7676558 Report

Quoted By:

>>7676473
please, do not make fun of us local ai artists ;(

Anonymous

Anonymous Mon 27 Nov 2023 01:58:44 No.7676559 Report

Quoted By:

>>7676541
ive been saying for months that the solution to the open source glut is tagging,
the focus should be in creating in a good dataset to train a good tagging mode, something like llava that is already an llm and whos text portion can then be reffited to a gen model.

Anonymous

Anonymous Mon 27 Nov 2023 01:59:25 No.7676560 Report

Quoted By: >>7676569 >>7676599

anyone else curious to see if nai embedded destructive patterns into their gens?

Anonymous

View Same Google ImgOps iqdb SauceNAO K4.png, 487KiB, 512x768

Anonymous Mon 27 Nov 2023 02:02:34 No.7676566 Report

Quoted By: >>7676611

>>7676554
Cheers, anon, thanks in advance for indulging me. One favor, though, if you can - steer away from huge tits? Not really my thing.

Anonymous

Anonymous Mon 27 Nov 2023 02:04:01 No.7676569 Report

Quoted By: >>7676577

>>7676560
"destructive patterns" in the sense of "data that makes resulting images unable to be trained", or what?

how does that shit even work, anyway? I get mislabeling images can fuck up the tagging/training process but I don't get how images themselves can be changed in such a way to prevent them from being trained on or to "pollute" training

Anonymous

Anonymous Mon 27 Nov 2023 02:04:40 No.7676571 Report

Quoted By:

>>7676557
Had a similar experience when trying out bladed weapons
Katana came out accurately almost 100% of the time
Kunai was a shit show

Anonymous

Anonymous Mon 27 Nov 2023 02:08:32 No.7676577 Report

Quoted By: >>7676587

>>7676569
>but I don't get how images themselves can be changed in such a way to prevent them from being trained on or to "pollute" training
there was a paper a while back that showed that can be done, i think this led to a service that does this for artists.

Anonymous

Anonymous Mon 27 Nov 2023 02:13:31 No.7676587 Report

Quoted By:

>>7676577
if you're talking about "glazing" then that shit is one denoise away from being usable lol, would have to do it for an extremely large dataset after a certain point in time though

Anonymous

View Same Google ImgOps iqdb SauceNAO 1700783338764923.png, 1MiB, 1216x1216

Anonymous Mon 27 Nov 2023 02:17:02 No.7676592 Report

Quoted By: >>7676594 >>7676595 >>7676597 >>7676598 >>7677275

Is NovelAI better than Stable Diffusion? I saw a thread that had really good mimicry of thiccwithaq's art style? I don't wanna pay 25$ for NovelAI but it looks pretty good tho

Anonymous

Anonymous Mon 27 Nov 2023 02:18:56 No.7676594 Report

Quoted By:

>>7676592
it's better
i'm not paying 25$ though

Anonymous

Anonymous Mon 27 Nov 2023 02:19:08 No.7676595 Report

Quoted By:

>>7676592
kinda, but we are still in the honeymoon phase

Anonymous

Anonymous Mon 27 Nov 2023 02:19:48 No.7676597 Report

Quoted By:

>>7676592
yes

Anonymous

View Same Google ImgOps iqdb SauceNAO 00124-160333519.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 02:19:51 No.7676598 Report

Quoted By: >>7676602 >>7676607

>>7676592
NAI uses SDXL, but finetuned and closed source.

That closed source for now it is better to mimic styles and following prompts. Also knows a lot more than local models.

Anonymous

Anonymous Mon 27 Nov 2023 02:19:52 No.7676599 Report

Quoted By: >>7676607 >>7676620

>>7676560
honestly with how horrible the eyes are I'm half tempted to think that they did it intentionally for it to be not "too good" and to fuck up anything trying to train with gens from it(not that it would be a good idea to do that anyway).
But they also probably just overtuned the fuck out of everything and it just happens that eyes can be the first/most noticeable thing to go when that happens.

Anonymous

Anonymous Mon 27 Nov 2023 02:22:37 No.7676602 Report

Quoted By: >>7676604

>>7676598
>NAI uses SDXL
bullshit

Anonymous

Anonymous Mon 27 Nov 2023 02:23:17 No.7676604 Report

Quoted By:

>>7676602
Are you being dense on purpose?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1700897001056963.png, 1020KiB, 1216x832

Anonymous Mon 27 Nov 2023 02:25:18 No.7676607 Report

Quoted By: >>7676610

>>7676598
>>7676599
How hard would it be to generate stuff like this on NovelAI? Do I have to tinker with prompts for hours or no? This really looks like a near perfect copy of the artstyle. (Also excuse the faggoty image I'm just using it for the artstyle)

Anonymous

Anonymous Mon 27 Nov 2023 02:26:58 No.7676610 Report

Quoted By:

>>7676607
extremely easy

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, 1boy, solo focus, Yoshiza (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 02:29:16 No.7676611 Report

Quoted By: >>7678379

>>7676566
That's fine. V3 has no real issue with her and understands her smaller details as well. PT outfit is also really easy to generate. You can even prompt in Soejima, but the generations I've seen with him really aren't that great.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1685898577243233.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 02:38:19 No.7676618 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 02:40:32 No.7676620 Report

Quoted By:

>>7676599
skill issue also meds

Anonymous

Anonymous Mon 27 Nov 2023 02:46:27 No.7676638 Report

Quoted By: >>7676671

i am seeing some weirdness in nai3, louise in particular has weird thick lines that stick out,
this in particular sticks out to me because that was a specific issue I had when I trained a louise lora months back.

Anonymous

Anonymous Mon 27 Nov 2023 02:50:02 No.7676641 Report

Quoted By:

>>7676480
almost got a good redrop style but not quite
still better than most lora i have seen

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, Yoshizawa Kasumi, persona (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 03:02:46 No.7676667 Report

Quoted By: >>7678379

Anonymous

View Same Google ImgOps iqdb SauceNAO fa703505-0bd8-4222-b40b-04024ed4 (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 03:04:58 No.7676671 Report

Quoted By: >>7676686 >>7676713

>>7676638
like this?

Anonymous

Anonymous Mon 27 Nov 2023 03:06:30 No.7676672 Report

Quoted By: >>7676678 >>7676679 >>7676682 >>7676696 >>7676707 >>7676726 >>7676736 >>7676745 >>7676749 >>7677399

Man why the fuck is Japan so bad at using NAI v3. Literally everything I'm seeing on Twitter is just "OMG IT CAN DO MY FAVORITE CHARACTER!!!" in either the anime's artstyle or the base artstyle of NAI v3 which looks like ass. Saw one dude actually unironically using the ai-generated tag too.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, Yoshizawa Kasumi, persona (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 03:07:36 No.7676675 Report

Quoted By:

>Switch from SFW artist to NSFW artist and the erotic part improves.
I guess you can only brute force a SFW artist so much into NSFW.

Anonymous

View Same Google ImgOps iqdb SauceNAO 64a7ff84-b09d-4f75-9812-3f593e01 (...).png, 2MiB, 1216x832

Anonymous Mon 27 Nov 2023 03:08:55 No.7676677 Report

Quoted By: >>7676686

might be an artist prompt

Anonymous

Anonymous Mon 27 Nov 2023 03:08:59 No.7676678 Report

Quoted By:

>>7676672
It's new tech, people will get good at it with time

Anonymous

Anonymous Mon 27 Nov 2023 03:09:00 No.7676679 Report

Quoted By:

>>7676672
>he unironically uses twitter

Anonymous

Anonymous Mon 27 Nov 2023 03:11:02 No.7676682 Report

Quoted By:

>>7676672
>Saw one dude actually unironically using the ai-generated tag too
my fucking sides

Anonymous

Anonymous Mon 27 Nov 2023 03:11:58 No.7676684 Report

Quoted By: >>7676689 >>7676692 >>7676699 >>7676700

"Finally, the AI can generate my waifu!"
>has her get fucked by a nigger

Why are people like this?

Anonymous

View Same Google ImgOps iqdb SauceNAO 0448df76-eea6-4d1f-93a8-884bca52 (...).png, 2MiB, 1216x832

Anonymous Mon 27 Nov 2023 03:12:07 No.7676686 Report

Quoted By:

without the artist tag
>>7676671
>>7676677

Anonymous

Anonymous Mon 27 Nov 2023 03:13:09 No.7676689 Report

Quoted By: >>7676693 >>7676699

>>7676684
hentai tends to draw even the main characters with darker skin during sex for some reason lmao

Anonymous

Anonymous Mon 27 Nov 2023 03:14:10 No.7676691 Report

Quoted By:

any tips for for generating sex scenes on nai?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, vladilena millize, 1boy, (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 03:14:49 No.7676692 Report

Quoted By: >>7676695

>>7676684
I'm a degenerate.

Anonymous

Anonymous Mon 27 Nov 2023 03:14:57 No.7676693 Report

Quoted By: >>7676700

>>7676689
Darker skin doesn't mean a nigger at all

Anonymous

Anonymous Mon 27 Nov 2023 03:16:56 No.7676695 Report

Quoted By: >>7676697

>>7676692
>I'm a degenerate.
You're a cuckold.

Anonymous

Anonymous Mon 27 Nov 2023 03:17:43 No.7676696 Report

Quoted By:

>>7676672
>Saw one dude actually unironically using the ai-generated tag too.
maybe they think using it to LARP artists is too dishonorable

Anonymous

Anonymous Mon 27 Nov 2023 03:17:51 No.7676697 Report

Quoted By: >>7676708

>>7676695
did a black guy steal your gf

Anonymous

Anonymous Mon 27 Nov 2023 03:18:31 No.7676698 Report

Quoted By: >>7676702 >>7676703

Does NovelAI censor some tags or ban you for getting too kinky?

Anonymous

Anonymous Mon 27 Nov 2023 03:18:57 No.7676699 Report

Quoted By: >>7676704

>>7676689
>>7676684
Typically the female is pale skinned so the male being slightly darker skinned by comparison helps her "pop" and be more obviously the focus of the art
This is one of the most common porn artist techniques so that you don't have the appearance of two light skinned characters blending together

Anonymous

Anonymous Mon 27 Nov 2023 03:19:28 No.7676700 Report

Quoted By:

>>7676693
>>7676684
Sometimes i still get very dark skin without the tag, could also be prompt gacha.

otherwise
1. mc is insufferable
2. they themselves are black/self insert as the black guy
3. are cucks
4. have a bbc fetish

Anonymous

Anonymous Mon 27 Nov 2023 03:19:58 No.7676702 Report

Quoted By:

>>7676698
>Getting too kinky
I've been genning a ton of preggo porn and I'm still ok so far

Anonymous

Anonymous Mon 27 Nov 2023 03:20:26 No.7676703 Report

Quoted By: >>7676709

>>7676698
NAI's whole thing is that they don't care what users are doing, but they aren't going to make it so a casual user can stumble upon NSFW.
It's why all of the NSFW tags do not auto-fill when typing them and why NSFW is part of the standard negative.

Anonymous

Anonymous Mon 27 Nov 2023 03:20:30 No.7676704 Report

Quoted By: >>7676710

>>7676699
i've never considered that, you learn something new every day

Anonymous

Anonymous Mon 27 Nov 2023 03:26:23 No.7676707 Report

Quoted By: >>7676724

>>7676672
japan is still using floppy disk and you are asking them to get good at AI

Anonymous

Anonymous Mon 27 Nov 2023 03:27:28 No.7676708 Report

Quoted By:

>>7676697
are they incapable of not stealing?

Anonymous

Anonymous Mon 27 Nov 2023 03:27:31 No.7676709 Report

Quoted By: >>7676711 >>7676714 >>7676716

>>7676703
Ah ok. Tho there are still certain tags that are way too risque and makes me nervous to put in. Maybe because NovelAI is a subscription service it just feels less private than running stable diffusion

Anonymous

Anonymous Mon 27 Nov 2023 03:27:53 No.7676710 Report

Quoted By:

>>7676704
In general, color contrast helps to draw your eyes. It's why black/red underwear, colored hair and colored eyes are all easy to remember

Anonymous

Anonymous Mon 27 Nov 2023 03:28:41 No.7676711 Report

Quoted By: >>7676715

>>7676709
One of the founding principals of NAI was protecting user privacy since the founding members were all people who got burned in the AI Dungeon shitshow.

Anonymous

Anonymous Mon 27 Nov 2023 03:29:56 No.7676713 Report

Quoted By:

>>7676671
no, using louise francoise le blanc de la valliere , zero no tsukaima, pink hair, long hair, flat chest, official art, anime screen cap
to get close to anime on style.

Anonymous

Anonymous Mon 27 Nov 2023 03:29:58 No.7676714 Report

Quoted By: >>7676719

>>7676709
I generated pregnant fizrot lolis all day you'll be fine

Anonymous

Anonymous Mon 27 Nov 2023 03:30:26 No.7676715 Report

Quoted By: >>7676731

>>7676711
>AI Dungeon shitshow.
tell me about this

Anonymous

Anonymous Mon 27 Nov 2023 03:30:59 No.7676716 Report

Quoted By: >>7676722

>>7676709
same granted im paying with it with gift cards i pay homeless people to get for me.

Anonymous

Anonymous Mon 27 Nov 2023 03:35:28 No.7676719 Report

Quoted By:

>>7676714
Proof? (catbox please)

Anonymous

Anonymous Mon 27 Nov 2023 03:38:57 No.7676722 Report

Quoted By:

>>7676716
Why are you concerned about payment tracking when they can just track you via your account

Anonymous

Anonymous Mon 27 Nov 2023 03:39:17 No.7676724 Report

Quoted By: >>7676733 >>7676749

>>7676707
More like NAI requires you to prompt entirely in English which filters them hard.

Anonymous

View Same Google ImgOps iqdb SauceNAO F_6DO_-boAEayPw[1].jpg, 219KiB, 832x1216

Anonymous Mon 27 Nov 2023 03:41:19 No.7676726 Report

Quoted By: >>7677253

>>7676672
https://twitter.com/takuuuuuuu1112
Are you talking about this guy

Anonymous

Anonymous Mon 27 Nov 2023 03:43:50 No.7676731 Report

Quoted By:

>>7676715
There was multiple things that I cannot even fully remember anymore, but one of the bigger shitshows was a combination of userdata being leaked and then it being found out that AI Dungeon was providing user written stories/prompts, without the user's consent, to a third party service who would review those stories because they had hit some kind of flag on AI Dungeon's side that marked them for review. The company that AI Dungeon had hired then also had a breach which led to the stories being leaked out onto the web. I think the stories were also being fed back to OpenAI who used that to form the basis of their censorship filter.
Pretty much AI Dungeon had absolutely no respect for user privacy of any kind and NAI was essentially born from the ashes of the entire AI Dungeon shitshow with one of their founding principals being to protect user privacy as much as they can.

Anonymous

Anonymous Mon 27 Nov 2023 03:44:26 No.7676733 Report

Quoted By: >>7676749

>>7676724
yeah well the fact that they suck at english is also linked to them still living in the 2000

Anonymous

Anonymous Mon 27 Nov 2023 03:44:35 No.7676734 Report

Quoted By: >>7676735

>1.3 Ownership. You, whether a legal or physical entity, retain all rights and ownership of your Content. We do not claim any ownership rights to your Content. Unless you agree and Anlatan agrees specifically to transfer your ownership to Anlatan.

Bros....I can make assets for a videogame......

Anonymous

Anonymous Mon 27 Nov 2023 03:45:14 No.7676735 Report

Quoted By: >>7676739 >>7676744

>>7676734
gabe will bonk you if you try to sell it tho

Anonymous

Anonymous Mon 27 Nov 2023 03:45:17 No.7676736 Report

Quoted By: >>7676749

>>7676672
I'm gonna say its the language barrier with the tags being in english and all that, but im just and ignorant ass American

Anonymous

Anonymous Mon 27 Nov 2023 03:49:29 No.7676739 Report

Quoted By: >>7676744

>>7676735
Fuck this gay earth

Anonymous

Anonymous Mon 27 Nov 2023 03:55:22 No.7676744 Report

Quoted By: >>7676769

>>7676735
>>7676739
Gabe doesn't need to know

Anonymous

Anonymous Mon 27 Nov 2023 03:56:24 No.7676745 Report

Quoted By: >>7676749

>>7676672
Japan isn't super great with AI image gen to begin with.
It's got nothing to do with "honor" or "respect," it's just a matter of art being a bigger deal in japan than it is in the west so most people that are actually into it actually try and pursue it(and anyone else/those who fail then fall into the salaryman pit where they have no time to dick with this shit).
And NAI isn't very compatible with people who are actually artistically inclined. No controlnet and no lora support makes it pretty worthless beyond some anatomy/some detail work but those aren't worth much since they can be filled in/corrected by hand anyway.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00311-484533329.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 03:59:40 No.7676747 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 04:03:11 No.7676748 Report

Quoted By: >>7676750

With novelAI can you generate......you know.....

Anonymous

View Same Google ImgOps iqdb SauceNAO 113741984_p0.png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 04:03:34 No.7676749 Report

Quoted By: >>7676752 >>7676753 >>7676755

>>7676672
>>7676736
>>7676733
>>7676724
>>7676745
https://www.pixiv.net/en/artworks/113741984
Well let's take a peek at some of these prompts shall we?
>{{{{Rinpoo, Breath of Fire2}}}}, {{art}}, {{beautiful}}, {{{{{{{{game cg}}}}}}}}, dynamic angle, dramatic pose, {{{{{{{masterpiece}}}}}}}, {{{{{{{best quality}}}}}}}, {{photo realistic}}, {{super detailed skin}}, {official art}, red short hair, {18 years old}, large breasts, cleavage, fighting stance, {{{{anime keyvisual}}}}, {{{{{cute}}}}},
>{{{{{{{Yumi, Senran Kagura}}}}}}}, {{art}}, {{beautiful}}, {{{{{{game cg}}}}}}, dynamic angle, dramatic pose, {{{{{{{masterpiece}}}}}}}, {{{{{{{best quality}}}}}}}, {{photo realistic}}, {{super detailed skin}}, {official art}, {18 years old}, {{cute}}, large breasts, panties, fighting stance, {{{{{{{2010s_(style)}}}}}}, {{{{{{{{{anime keyvisual}}}}}}}}},
>{{{{{Lenna Charlotte Tycoon, FINAL FANTASY}}}}}, {{art}}, {{beautiful}}, {{{{{{game cg}}}}}}, dynamic angle, dramatic pose, {18 years old}, {{cute}}, large breasts, cleavage, [[white panties]], {bare legs}, {{2010s_(style)}}, {{anime keyvisual}},

Anonymous

Anonymous Mon 27 Nov 2023 04:04:21 No.7676750 Report

Quoted By:

>>7676748
YJK?
Yes.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00012-3754422040.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 04:04:58 No.7676751 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 04:06:20 No.7676752 Report

Quoted By:

>>7676749
nips are retarded

Anonymous

Anonymous Mon 27 Nov 2023 04:07:56 No.7676753 Report

Quoted By:

>>7676749
we need a schizo neg prompt 2.0, that guy was great.

Anonymous

Anonymous Mon 27 Nov 2023 04:09:54 No.7676755 Report

Quoted By:

>>7676749
>nips still 1.2 prompting
Based and super detailed Rutkowskipilled.

Anonymous

View Same Google ImgOps iqdb SauceNAO 16598.png, 4MiB, 1536x1536

Anonymous Mon 27 Nov 2023 04:10:36 No.7676758 Report

Quoted By: >>7676759 >>7676767 >>7676885

Just went with it and got the 25 subscription. Solid upgrade over local, while not mind blowing, i am already able to do shit (yuri finally, actually decent full body with decent feet/hands) I struggled with.

Do parenthesis and example (subject:1.3) still work the same way for emphasis like local?

What the fuck are { } these things doing?

BREAK still a thing?

Masterpeice, hi res,... still necessary?

Anonymous

Anonymous Mon 27 Nov 2023 04:14:59 No.7676759 Report

Quoted By: >>7676765 >>7676885 >>7677007

>>7676758
{ } are the replacement for parentheses after Voldy told them to fuck off
BREAK was never implemented in NAI
NAI stopped using MASTERPIECE and started using VERY AESTHETIC to tag their datasets V2 and onwards.
Using hi res in your prompt never made sense from the start since those images are downscaled before being trained on anyways.

Anonymous

View Same Google ImgOps iqdb SauceNAO 16719.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 04:28:22 No.7676765 Report

Quoted By:

>>7676759

Thanks anon,

I'm mostly importing most of my prompts/what worked for me from local. Any prompt composition you could recommend or is it all up in the air still?

Anonymous

Anonymous Mon 27 Nov 2023 04:28:30 No.7676767 Report

Quoted By: >>7676885

>>7676758
https://docs.novelai.net/image/qualitytags.html and the next one for undesired content are worth a quick read.
@_@ in negatives for eyes and there's probably a bunch of tags like that we still didn't figure out.
>masterpiece etc
best quality, very aesthetic if you really want to but it's added automatically unless you turn it off. "ai-generated" is apparently also a tag, hard to tell if it can be leveraged someway

Anonymous

Anonymous Mon 27 Nov 2023 04:33:22 No.7676769 Report

Quoted By: >>7676771 >>7676776 >>7676820 >>7677350

>>7676744
As far as I understand if the steam employee reviewing your game suspects that you used ai for the artwork he will ask you for WIPs, and if you fail to provide them your game will be rejected.

Anonymous

Anonymous Mon 27 Nov 2023 04:34:34 No.7676771 Report

Quoted By:

>>7676769
Bullshit

Anonymous

View Same Google ImgOps iqdb SauceNAO DEEP HURTING, nsfw,1girl,1monkey (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 04:37:30 No.7676775 Report

Quoted By: >>7676778 >>7676779

It feels kinda crazy how you can put in something insane like 'fingered by a monkey', and get completely coherent results on your first result.

Anonymous

Anonymous Mon 27 Nov 2023 04:37:39 No.7676776 Report

Quoted By: >>7676781 >>7676820 >>7677350

>>7676769
So don't get suspected 5head
Not like that's hard with NAI v3. Or 1.5 with loras if you really know what you're doing and manage to scrub out all the tells.

Anonymous

Anonymous Mon 27 Nov 2023 04:38:34 No.7676778 Report

Quoted By:

>>7676775
It's the SDXL side of it that likes natural language over prompts. You can brute force certain things through natural language, but natural language will eat more tokens.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00018-270469421.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 04:39:42 No.7676779 Report

Quoted By:

>>7676775
I like that anal cowgirl finally does anal and not vaginal

Anonymous

Anonymous Mon 27 Nov 2023 04:42:04 No.7676781 Report

Quoted By:

>>7676776
>if you really know what you're doing and manage to scrub out all the tells.
easier said than done

Anonymous

Anonymous Mon 27 Nov 2023 04:46:25 No.7676786 Report

Quoted By: >>7676789 >>7676791 >>7676804

So... How's this work, you fork over $25 to the NAI guys and you get a certain amount of tokens a day or week to spend, or as many as you can muster? From what I'm seeing, looks like someone's managed to integrate it into automatic1111, so you can even refine the results through inpainting and img2img?

Anonymous

Anonymous Mon 27 Nov 2023 04:47:48 No.7676789 Report

Quoted By: >>7676794

>>7676786
You spend 25 bucks and you get access to unlimited image generation at their medium size image and below.

Anonymous

Anonymous Mon 27 Nov 2023 04:48:04 No.7676791 Report

Quoted By: >>7676792 >>7676794 >>7676801

>>7676786
You get 10k tokens and 30 days of unlimited gens at below 2048 combined pixels

Anonymous

Anonymous Mon 27 Nov 2023 04:51:38 No.7676792 Report

Quoted By: >>7676793 >>7676799 >>7676801

>>7676791
>2048 pixels
???? that's nothing

Anonymous

Anonymous Mon 27 Nov 2023 04:53:18 No.7676793 Report

Quoted By:

>>7676792
If you want to go above that it costs like 10 "credits" per image to 1.5x upscale, more if you want to go to increasingly higher resolutions.

Anonymous

Anonymous Mon 27 Nov 2023 04:53:37 No.7676794 Report

Quoted By: >>7676798 >>7676801

>>7676789
>>7676791
So is it unlimited, or token based? I see lots of people mention tokens, is 10k just high enough that it's effectively unlimited or are there different tiers to get? Also, I'm assuming monthly payment, right?

Anonymous

Anonymous Mon 27 Nov 2023 04:55:33 No.7676798 Report

Quoted By: >>7676801

>>7676794
225 tokens per positive/negative prompt, using more than that I think it starts to using the currency that nai uses

Anonymous

Anonymous Mon 27 Nov 2023 04:56:34 No.7676799 Report

Quoted By:

>>7676792
you can just upscale locally

Anonymous

View Same Google ImgOps iqdb SauceNAO 9dba773b-5866-4724-85cc-72f13289 (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 05:00:55 No.7676801 Report

Quoted By: >>7676803

>>7676791
>>7676792
The max is 1024x1024, or 1048576 pixels. This is actually the target resolution for the model anyway, and you get diminishing returns going higher.

>>7676794
Both, below the limits things are free, otherwise you pay. 10000 tokens is less than 400 images, I generated that many today alone, paying for tokens is a total rip off.

>>7676798
No it doesn't cost anything, it just ignores anything beyond 225.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00022-2080730661.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 05:02:29 No.7676803 Report

Quoted By: >>7676806

>>7676801
>No it doesn't cost anything, it just ignores anything beyond 225.
oh mb, thanks for the correction.

is there a way to get good eyes that doesn't involve genning the character real close?

Anonymous

Anonymous Mon 27 Nov 2023 05:02:44 No.7676804 Report

Quoted By:

>>7676786
you could always inpaint and img2img, the auto is just an UI change

Anonymous

Anonymous Mon 27 Nov 2023 05:06:06 No.7676806 Report

Quoted By:

>>7676803
@_@ in negatives. Change artists if it persists

Anonymous

Anonymous Mon 27 Nov 2023 05:20:25 No.7676817 Report

Quoted By: >>7676819

Let's say I get access to an Opus API key but not the actual account. Will I be able to use imagen somehow, like through an extension or something? I know for a fact I could use textgen through SillyTavern this way.

Anonymous

Anonymous Mon 27 Nov 2023 05:22:08 No.7676819 Report

Quoted By: >>7676835

>>7676817
yes you can use imagen through api

Anonymous

Anonymous Mon 27 Nov 2023 05:22:21 No.7676820 Report

Quoted By: >>7676831

>>7676769
>>7676776
What if you sell your game on itch.io? Are they still gonna check that the assets are not AI?

Anonymous

Anonymous Mon 27 Nov 2023 05:26:16 No.7676831 Report

Quoted By:

>>7676820
no they don't care, but the userbase there is tiny

Anonymous

Anonymous Mon 27 Nov 2023 05:28:09 No.7676835 Report

Quoted By: >>7676839

>>7676819
Ok, what's the extension to use it through the webui? Or is there a different repo for that entirely?

Anonymous

Anonymous Mon 27 Nov 2023 05:31:58 No.7676839 Report

Quoted By: >>7676852

>>7676835
ctrl+f github

Anonymous

Anonymous Mon 27 Nov 2023 05:44:03 No.7676852 Report

Quoted By: >>7676854 >>7676871

>>7676839
All I could find were a couple of extremely barebones WIP repos and a chink program without a translation.

Anonymous

Anonymous Mon 27 Nov 2023 05:47:29 No.7676854 Report

Quoted By: >>7676855

>>7676852
he means this
https://github.com/Metachs/sdwebui-nai-api

Anonymous

Anonymous Mon 27 Nov 2023 05:48:58 No.7676855 Report

Quoted By:

>>7676854
Thanks anon, no wonder I didn't find it, I was searching for "novelai" because "nai" seemed a bit too broad of a request after a quick test search.

Anonymous

Anonymous Mon 27 Nov 2023 05:53:09 No.7676858 Report

Quoted By: >>7676891

How do you change the denoise when inpainting on nai?

Anonymous

Anonymous Mon 27 Nov 2023 05:53:32 No.7676860 Report

Quoted By:

>>7676411
weird, this isnt AOM
>>7676473
when NAI leaked you raged? why your blood pressure is high over ai-generated images?

Anonymous

Anonymous Mon 27 Nov 2023 06:04:01 No.7676871 Report

Quoted By:

>>7676852
I was telling you to ctrl+f "github" in the thread... How are you smart enough to scrape API keys but too dumb to understand this?

Anonymous

Anonymous Mon 27 Nov 2023 06:07:15 No.7676874 Report

Quoted By: >>7676877 >>7676878

if there's no limit on nai normal generation why aren't there public proxies of api keys just like at /aicg/. are you guys that petty?

Anonymous

Anonymous Mon 27 Nov 2023 06:09:13 No.7676877 Report

Quoted By:

>>7676874
api usage like get banned immediately that why

Anonymous

Anonymous Mon 27 Nov 2023 06:10:16 No.7676878 Report

Quoted By:

>>7676874
retard, NAI keys are omega niche
or are you expecting people to share their own and get their accounts banned?
>/aicg/
Only Claude proxies are there, even though GPT4 keys are retardedly easy to find

Anonymous

Anonymous Mon 27 Nov 2023 06:15:38 No.7676885 Report

Quoted By:

>>7676758
>>7676759
>>7676767
Where can I get a quick rundown on the parentheses and commas? I don't see it any of the OP links and I'm new to this.

Anonymous

Anonymous Mon 27 Nov 2023 06:25:18 No.7676891 Report

Quoted By: >>7676901

>>7676858
You can't, NAI only offers the option to generate from scratch.

Anonymous

Anonymous Mon 27 Nov 2023 06:40:19 No.7676901 Report

Quoted By: >>7676902

>>7676891
You can inpaint and img2img on NAI but no denoise slider

Anonymous

Anonymous Mon 27 Nov 2023 06:40:52 No.7676902 Report

Quoted By:

>>7676901
There is one for img2img but dunno about inpaints

Anonymous

View Same Google ImgOps iqdb SauceNAO tmpfzik6w9m.png, 336KiB, 512x768

Anonymous Mon 27 Nov 2023 06:50:40 No.7676903 Report

Quoted By:

Finding pictures to train for this position is a pain

Anonymous

View Same Google ImgOps iqdb SauceNAO amazing maybe nai3.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 07:04:13 No.7676910 Report

Quoted By: >>7678379

>>7676544
maybe

Anonymous

View Same Google ImgOps iqdb SauceNAO ComfyUI_00019_.png, 957KiB, 920x920

Anonymous Mon 27 Nov 2023 07:16:35 No.7676918 Report

Quoted By: >>7676921

>>7676332
>>7676440
Wanted to see if I could keep my gens in a fappable format, and just img2img them before posting on /h/. Replace all the quality tags with "soul" crap like retro artstyle or limited color palette.

Anonymous

Anonymous Mon 27 Nov 2023 07:18:34 No.7676921 Report

Quoted By: >>7676933

>>7676918
>tags:soul,soulful
>negs:soulless,slop

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1boy, 1girl, mitarashi ank (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 07:23:01 No.7676923 Report

Quoted By: >>7676926 >>7676928 >>7676987 >>7676992 >>7677347

whats that one huggingface thats img2prompt

Anonymous

Anonymous Mon 27 Nov 2023 07:28:42 No.7676926 Report

Quoted By:

>>7676923
Fuck, really nice face Anon. Way to go.

Anonymous

View Same Google ImgOps iqdb SauceNAO { {{{character_sumeragi erika (s (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 07:29:26 No.7676927 Report

Quoted By: >>7677046 >>7677735

For anyone else fucking around with NAIv3, I tested a little over a hundred or so different artists in NAIv3 on mostly the same prompt for my own amusement, but maybe you'll find it interesting.

If interested:
https://www.pixiv.net/en/artworks/113755446
https://www.pixiv.net/en/artworks/113755532

Direct catboxes:
https://files.catbox.moe/ivf0ep.zip
https://files.catbox.moe/z3lnus.zip

Anonymous

Anonymous Mon 27 Nov 2023 07:31:12 No.7676928 Report

Quoted By: >>7676930 >>7676936 >>7676992 >>7677012 >>7678005

>>7676923
I just got NovelAI, how do you get that level of quality? Are those in the filename really the only tags used? Or is this like a luck-based thing

Anonymous

Anonymous Mon 27 Nov 2023 07:32:00 No.7676930 Report

Quoted By: >>7676931 >>7676936

>>7676928
are you telling us the base quality is worse than that

Anonymous

Anonymous Mon 27 Nov 2023 07:37:12 No.7676931 Report

Quoted By: >>7676934

>>7676930
Compared to amslop this is eternal masterpiece

Anonymous

Anonymous Mon 27 Nov 2023 07:38:23 No.7676932 Report

Quoted By: >>7677042 >>7677604

Am I missing something here? When I import to png info I don't anything denoting the position just character info and nsfw
https://files.catbox.moe/iq65tb.jpg

Anonymous

Anonymous Mon 27 Nov 2023 07:39:07 No.7676933 Report

Quoted By:

>>7676921
This is the future

Anonymous

Anonymous Mon 27 Nov 2023 07:39:27 No.7676934 Report

Quoted By:

>>7676931
enjoy sniffing your own asshole? check the shadow on top of the head, that is slop

Anonymous

View Same Google ImgOps iqdb SauceNAO Anon Never Lies.png, 19KiB, 1297x272

Anonymous Mon 27 Nov 2023 07:41:54 No.7676936 Report

Quoted By: >>7676944 >>7676945 >>7676972 >>7677157 >>7677281

>>7676928
No, you can get the full tags by loading the catbox image into NAI directly or using something like https://novelai-metadata-viewer.pages.dev/

>>7676930
The base quality of NAIv3 is complete trash, even if you use their recommended quality tags. To be fair, this is not entirely their fault: 90% of the art uploaded on danbooru is trash, and the model accurately replicates this.

What is their fault is the fact that they meta tagged shit for quality but don't even explain how they did it even on their own mini-wiki.

For example, danbooru scores were transcribed during training into the following categories:
best quality
amazing quality
great quality
normal quality
bad quality
worst quality

NAI further did some voodoo that they refuse to elaborate on and applied their own scoring using this following tags:
very aesthetic
aesthetic
displeasing
very displeasing

It is very important to use these tags in the positive and negative. ALSO, according to the supreme authority on such matter - anonymous chan posters - they apparently trained with quality tags at the end and recommend putting them at the end for that reason instead of the beginning. I've tested this and it matters a lot.

FURTHERMORE, characters are actually inputted during training as character:charactername and artists as artist:artistname. All of this information is absolutely essential to get NAIv3 to not be a flaming garbage pile but NONE OF IT is on their own help pages. Very, very bad.

That said, despite all NAI's bullshit, you can get some pretty neat stuff out of it anyway. I'm just fucking around with it and it's definitely interesting, although probably without as much staying power as local since it won't improve as fast.

Anonymous

Anonymous Mon 27 Nov 2023 07:54:31 No.7676944 Report

Quoted By:

>>7676936
I mean this how the local furry model operate as well without artist tags base gen are absolute crap but it gives way better variety instead of the almagnation of artist being into one like v1 NAI which stripped artist tags.
if your not using artist tag you should be using some form the (medium) tags like oil painting (medium) or lineart to push it towards something again like the furry model for local.

Anonymous

Anonymous Mon 27 Nov 2023 07:55:12 No.7676945 Report

Quoted By:

>>7676936
Nice find anon
Curious why they dont explain how to use their own model

Anonymous

Anonymous Mon 27 Nov 2023 07:58:40 No.7676949 Report

Quoted By: >>7676955 >>7676962 >>7676967

Can any of the local lovers show examples of good-looing spread pussy and spread pussy under clothes gens? Yes, I am having a skill issue, it seems because it always looked like shit when I tried it.

Anonymous

Anonymous Mon 27 Nov 2023 08:02:42 No.7676955 Report

Quoted By: >>7676956

>>7676949
More or less massive gamble, 1.5 with loras really struggles with this concept
https://archiveofsins.com/h/thread/7574428/#7574853

Anonymous

Anonymous Mon 27 Nov 2023 08:04:31 No.7676956 Report

Quoted By:

>>7676955
I see, thank you

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, blonde hair, blunt bangs, (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 08:07:50 No.7676959 Report

Quoted By: >>7676963 >>7677044 >>7677090

Well I just finished going down a rabbit hole of testing out various VN artists and was shocked that I was able to get results even if there was only a 150 or so images reported on Danbooru.

Anonymous

View Same Google ImgOps iqdb SauceNAO ComfyUI_00006_.png, 905KiB, 920x920

Anonymous Mon 27 Nov 2023 08:11:14 No.7676962 Report

Quoted By: >>7677119

>>7676949
The fingers spreading it are the hard part. As long as it's not upside down, then the model breaks and loras won't help.

Haven't tried under clothes yet, I'll see what it does. Might be easier to make the pussy normally first, then inpaint clothes with see-through tag.

Anonymous

Anonymous Mon 27 Nov 2023 08:12:15 No.7676963 Report

Quoted By: >>7676966 >>7677090

>>7676959
yea I've gotten results with about 100 and it not counting the deleted stuff either

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{{isabelle (shadowverse)}}}}, (...).png, 2MiB, 1536x1536

Anonymous Mon 27 Nov 2023 08:18:12 No.7676966 Report

Quoted By: >>7676967 >>7676968 >>7676969

>>7676963
V3 seems to have taken the minimum examples for basic coherence from like 500 in v1 to 50ish. I don't know how they did it but it is impressive as shit.

You do need a few supplementary tags for nailing details when the number of training examples it low though. Just a few tags that show up a lot with the target tag help to guide it in the right direction.

This character, tagged Isabelle (shadowverse), has 108 tagged images on danbooru. I'd estimate over half of those are noncanon costumes or nude or close-ups that don't show the whole costume. The fact that NAIv3 can get the costume this good, even with costume tags helping, is actually some fucked up voodoo shit. Skynet soon, brothers.

Anonymous

View Same Google ImgOps iqdb SauceNAO 02457-6663776166.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 08:20:38 No.7676967 Report

Quoted By: >>7676980

>>7676949
i will try soon
>>7676966
now they need to find a way to boost the quality
>pic
figured a way to kill AOM traits (ribs and stuff), interesting how the picture becomes less generic

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, blonde hair, blunt bangs, (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 08:23:35 No.7676968 Report

Quoted By:

>>7676966
They definitely did something. I don't know if they managed to scrape art that was possibly hidden, but not deleted or something since I shouldn't be able to get isako rokurou since Danbooru reports only having 84 images tied to the artist, but I didn't even need to strengthen the artist tag to see that it was already shifting towards Rokurou's style.

Anonymous

Anonymous Mon 27 Nov 2023 08:24:54 No.7676969 Report

Quoted By:

>>7676966
yea they talked about improving the result from low image count in the training and yea as long you nudge in a few reminder tags like the series of said character it picks up on it instantly after

Anonymous

View Same Google ImgOps iqdb SauceNAO asooka.png, 331KiB, 512x512

Anonymous Mon 27 Nov 2023 08:27:32 No.7676972 Report

Quoted By: >>7676979 >>7676985

>>7676936
local has been stagnant for the past year

Anonymous

View Same Google ImgOps iqdb SauceNAO HexCowPaizuri sUHD.png, 3MiB, 1229x1843

Anonymous Mon 27 Nov 2023 08:29:33 No.7676973 Report

Quoted By: >>7677347

>Forgot to inpaint the milk label properly on the milk bottle despite all the other autistic inpaints like the cow and the cervix punching

Anonymous

Anonymous Mon 27 Nov 2023 08:34:08 No.7676979 Report

Quoted By: >>7677204

>>7676972
Hard disagree. In the last year we've had: custom model mixes and fine tunes, loras and TIs, controlnet, and tons of tweak extensions like FreeU and kohya's hires thing.

But yeah, there's only so far all this can take us when it's built on top of an old model. Though, the furries have definitely demonstrated that all the SD 1.4/1.5 based anime models are punching WAY below what they would be capable of if we had a decent fine tune with modern methods and no artist pruning.

Why the fuck do furries have so much money? I remember I once saw one who bought a fucking tank. A TANK. Give me a second: https://twitter.com/XavierFox19/status/1230541678693539842

Anonymous

View Same Google ImgOps iqdb SauceNAO 02815-4479949907.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 08:34:29 No.7676980 Report

Quoted By: >>7676998

>>7676967

Anonymous

Anonymous Mon 27 Nov 2023 08:35:23 No.7676982 Report

Quoted By: >>7676983 >>7677045 >>7677207

If I want to generate 2 characters in one image can I specify a different artstyle for both? Or will AI mesh together both artists if I name 2?

Anonymous

Anonymous Mon 27 Nov 2023 08:36:15 No.7676983 Report

Quoted By:

>>7676982
The latter.

The only way you could do it, would be to use something like latent couple to apply a lora/prompt to only part of an image, or to inpaint using separate images/prompts.

Anonymous

Anonymous Mon 27 Nov 2023 08:38:33 No.7676985 Report

Quoted By:

>>7676972
We've probably been fairly stagnant on the hard model side of things since feb or march or something.
The actual PROCESS and shit that you can do to improve things has been pretty drastic compared to then, though.
People are still fucking sleeping on 1:1 resolution high denoise img2img iteration with controlnet tile.

Anonymous

View Same Google ImgOps iqdb SauceNAO Aerithdongworshipalt.png, 693KiB, 832x1216

Anonymous Mon 27 Nov 2023 08:40:49 No.7676987 Report

Quoted By:

>>7676923
Just use the catbox browser addon. You can drop the images with metadata into NAI3

Anonymous

View Same Google ImgOps iqdb SauceNAO jhy.png, 475KiB, 576x960

Anonymous Mon 27 Nov 2023 08:41:53 No.7676989 Report

Quoted By: >>7677165

Can anyone try KH Aqua in v3? I feel that even with loras she never looks right

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{(narmaya (granblue fantasy)}} (...).png, 2MiB, 1536x1536

Anonymous Mon 27 Nov 2023 08:43:35 No.7676992 Report

Quoted By: >>7677321

>>7676928
>>7676923
Been fucking around with novelAI a lot the last few days.

I think the best route is to use the 0-cost 28-step prompt at around 5-6 guidance until you get a good base image. Then hit the enhance option and bump up the magnitude to like 4 or 5, and you can up the step count as well. It will produce a much higher quality image but can also randomly fuck itself up (controlled by the strength option in the advanced enhance settings). However a low strength makes it less likely to improve the overall image quality when enhancing. So overall enhance is like a gacha game whenever you roll so pray to the AI gods it doesn't fuck the image up.

You can get also pretty descriptive with the tags like "girl has left hand holding (object)" and it will work. Also another anon mentioned the quality tags, you need to add them to the positive and negative. Literally grab this image and throw it in for a bunch of added negative traits to remove.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, nagatoro hayase, ijiranai (...).png, 889KiB, 832x1216

Anonymous Mon 27 Nov 2023 08:47:40 No.7676997 Report

Quoted By: >>7677002

Anonymous

View Same Google ImgOps iqdb SauceNAO 01201-6521849941.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 08:49:14 No.7676998 Report

Quoted By: >>7677659

>>7676980
forgot lora

Anonymous

Anonymous Mon 27 Nov 2023 08:52:22 No.7677002 Report

Quoted By: >>7677008

>>7676997
GOD FUCKING DAMNIT
i've been trying to do spread pussy under clothes on local for at least 20 minutes without any fucking success.
IT'S FUCKING OVER

Anonymous

Anonymous Mon 27 Nov 2023 08:57:19 No.7677007 Report

Quoted By:

>>7676759
>{ } are the replacement for parentheses after Voldy told them to fuck off
They used {} from the start, they just reimplemented the same logic when he did.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, nagatoro hayase, ijiranai (...).png, 788KiB, 832x1216

Anonymous Mon 27 Nov 2023 08:58:21 No.7677008 Report

Quoted By: >>7677009 >>7677014

>>7677002
yeah that's what got me when I saw it do this shit from prompt alone. this was a fetish that I couldn't do on local and now it feels like I'm gonna coom myself to death

Anonymous

Anonymous Mon 27 Nov 2023 09:00:04 No.7677009 Report

Quoted By: >>7677010 >>7677017 >>7677070

>>7677008
can it do clothed (anal) penetration?

Anonymous

Anonymous Mon 27 Nov 2023 09:02:47 No.7677010 Report

Quoted By:

>>7677009
I'll look up the tags but probably not

Anonymous

View Same Google ImgOps iqdb SauceNAO 3ad79fa1-e275-412b-a51a-baa9fcbb (...).png, 985KiB, 832x1216

Anonymous Mon 27 Nov 2023 09:03:00 No.7677011 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 09:03:47 No.7677012 Report

Quoted By:

>>7676928
You can drop PNG images directly from 4chan into NAI to import the settings.

Anonymous

Anonymous Mon 27 Nov 2023 09:09:49 No.7677014 Report

Quoted By: >>7677016 >>7677017 >>7677018 >>7677020 >>7677031 >>7677032

>>7677008
What does "can't do on local" mean?

Anonymous

Anonymous Mon 27 Nov 2023 09:12:08 No.7677016 Report

Quoted By:

>>7677014
you literally can't get the concept generated with pure txt2img prompting, because model has no data of said concept.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, nagatoro hayase, ijiranai (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 09:12:09 No.7677017 Report

Quoted By: >>7677021

>>7677009
Yeah it kinda insists on adding cutout it seems, neg doesn't help
>>7677014
Means I couldn't make something I could fap to

Anonymous

Anonymous Mon 27 Nov 2023 09:12:12 No.7677018 Report

Quoted By: >>7677022 >>7677026

>>7677014
He was trying to say "I have a chronic skill issue".

Anonymous

Anonymous Mon 27 Nov 2023 09:13:17 No.7677020 Report

Quoted By:

>>7677014
It means that this is an extremely hard concept to prompt for on local (one your own pc, based on SD1.4/1.5) AI models.

Technically, SDXL is also local (open source). But it was just trained on LAION and whatever web shit they scraped - real photographs with captions. Not very useful for anime art. But what everyone is talking about when they are talking nonlocal right now is novelai's fine tune of SDXL on danbooru. It's a lot better at tag comprehension than previous anime models. A lot of things that are difficult to do and require fucking around with loras and controlnets to achieve on local models are just straightforwardly promptable on NAIv3, which is the real edge it has IMO.

Anonymous

Anonymous Mon 27 Nov 2023 09:13:24 No.7677021 Report

Quoted By: >>7677085

>>7677017
Just open photoshop, get a graphic tablet and draw it yourself fucking zoomer

Anonymous

Anonymous Mon 27 Nov 2023 09:13:26 No.7677022 Report

Quoted By: >>7677027 >>7677029 >>7677037 >>7677057

>>7677018
would like to see a single "skilled" person to do it with txt2img only, no loras

Anonymous

Anonymous Mon 27 Nov 2023 09:14:33 No.7677026 Report

Quoted By:

>>7677018
I didn't ask this in thread before for nothing. No one provided examples so I assume you're just having another meltie mate

Anonymous

Anonymous Mon 27 Nov 2023 09:15:54 No.7677027 Report

Quoted By:

>>7677022
You can't do it with inpainting and loras either. Maybe with canny controlnet of a real spread? but I'm gonna vomit from imagining what kind of monstrosities I'd have to sit through

Anonymous

Anonymous Mon 27 Nov 2023 09:21:48 No.7677029 Report

Quoted By: >>7677033

>>7677022
>txt2img only, no loras
why? loras are a must for local but the issue is that it you can't do something that doesn't look like trash for this concept

Anonymous

View Same Google ImgOps iqdb SauceNAO TotK_Queen_Sonia_Artwork.png, 2MiB, 1000x1747

Anonymous Mon 27 Nov 2023 09:22:17 No.7677030 Report

Quoted By:

can someone please make a lora for Queen Sonia from Tears of the Kingdom

Anonymous

Anonymous Mon 27 Nov 2023 09:22:52 No.7677031 Report

Quoted By:

>>7677014
a way to sell nai3
it is stupid because all this insulting local users will get local users to create loras from nai3 prompts

Anonymous

Anonymous Mon 27 Nov 2023 09:23:30 No.7677032 Report

Quoted By:

>>7677014
means that you don't need ai shit retard, just get a pencil and do it yourself god damn it

Anonymous

Anonymous Mon 27 Nov 2023 09:23:47 No.7677033 Report

Quoted By: >>7677041

>>7677029
loras should only exist for characters and styles since new only are coming out all the time, not concepts though. the perfect model needs to know all the concepts by itself.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00424-6485697104.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 09:26:26 No.7677037 Report

Quoted By: >>7677040

>>7677022
im sure if i kept going i could get full spread

Anonymous

Anonymous Mon 27 Nov 2023 09:29:24 No.7677039 Report

Quoted By:

Breathing Life Into Sketches Using Text-to-Video Priors
https://arxiv.org/abs/2311.13608
>A sketch is one of the most intuitive and versatile tools humans use to convey their ideas visually. An animated sketch opens another dimension to the expression of ideas and is widely used by designers for a variety of purposes. Animating sketches is a laborious process, requiring extensive experience and professional design skills. In this work, we present a method that automatically adds motion to a single-subject sketch (hence, "breathing life into it"), merely by providing a text prompt indicating the desired motion. The output is a short animation provided in vector representation, which can be easily edited. Our method does not require extensive training, but instead leverages the motion prior of a large pretrained text-to-video diffusion model using a score-distillation loss to guide the placement of strokes. To promote natural and smooth motion and to better preserve the sketch's appearance, we model the learned motion through two components. The first governs small local deformations and the second controls global affine transformations. Surprisingly, we find that even models that struggle to generate sketch videos on their own can still serve as a useful backbone for animating abstract representations.
https://livesketch.github.io
https://github.com/yael-vinker/live_sketch
bretty cool

Anonymous

Anonymous Mon 27 Nov 2023 09:29:25 No.7677040 Report

Quoted By:

>>7677037
just looks like hands on crotch, cameltoe.

Anonymous

Anonymous Mon 27 Nov 2023 09:31:45 No.7677041 Report

Quoted By: >>7677043

>>7677033
>the perfect model needs to know all the concepts by itself
yeah but there's no perfect model
nai seem to pick up concepts with 1k+ pics on danbooru (depending on quality of the dataset probably)
it can't do pantyjobs which is sad

Anonymous

Anonymous Mon 27 Nov 2023 09:34:45 No.7677042 Report

Quoted By: >>7677576

>>7676932
Update catbox script from OP and use that, NAI import doesn't work correctly

Anonymous

Anonymous Mon 27 Nov 2023 09:35:28 No.7677043 Report

Quoted By: >>7677048

>>7677041
does NAI have prompt scheduling? like [holding panties:0.4] in auto1111? i'm pretty sure if it was available you could make naiv3 be able to do pantyjobs with [handjob:holding panties:0.3] or something. it is far smarter and comprehensive that what we have locally.

Anonymous

Anonymous Mon 27 Nov 2023 09:37:26 No.7677044 Report

Quoted By: >>7677090

>>7676959
Yep. I was surprised to get VN charas with 30 images to work too

Anonymous

Anonymous Mon 27 Nov 2023 09:40:00 No.7677045 Report

Quoted By:

>>7676982
You can gen 2 characters and inpaint separately after

Anonymous

Anonymous Mon 27 Nov 2023 09:40:02 No.7677046 Report

Quoted By: >>7677049

>>7676927
Are you by chance that Erika anon from October 2022 that was trying to figure out inpainting then made a guide on it?

Anonymous

Anonymous Mon 27 Nov 2023 09:40:32 No.7677048 Report

Quoted By:

>>7677043
it has a "prompt mixing" feature
https://docs.novelai.net/image/promptmixing.html
I haven't experimented with it yet, guess I should try

Anonymous

Anonymous Mon 27 Nov 2023 09:40:48 No.7677049 Report

Quoted By: >>7677052

>>7677046
Yes.

Anonymous

Anonymous Mon 27 Nov 2023 09:42:28 No.7677052 Report

Quoted By: >>7677065

>>7677049
Cool that you are still around.

Anonymous

Anonymous Mon 27 Nov 2023 09:43:30 No.7677053 Report

Quoted By: >>7677054 >>7677055

Can we ban SDscum now that paychads are the factual superior kind?
Localniggers are the new Orangeshitmix.

Anonymous

Anonymous Mon 27 Nov 2023 09:44:12 No.7677054 Report

Quoted By:

>>7677053
fuck off with your pathetic false flag

Anonymous

Anonymous Mon 27 Nov 2023 09:44:59 No.7677055 Report

Quoted By:

>>7677053
Can u go post cheap bait on /g/ now? Gets tiring after a few days

Anonymous

View Same Google ImgOps iqdb SauceNAO 1699722650528477.png, 893KiB, 768x1152

Anonymous Mon 27 Nov 2023 09:45:54 No.7677057 Report

Quoted By: >>7677059 >>7677060 >>7677061 >>7677104 >>7677112 >>7677115 >>7677117 >>7677357 >>7677383

>>7677022
>txt2img only, no loras
Why should I gimp myself and discard local's trump cards just to please NAIshills? If I want to get a pic, I'll use everything at my disposal instead of artificially limiting my options like a prude retard. I'm not a scholar scientifically comparing base models, I'm an everyday user who just wants results without paypigging.

Anonymous

View Same Google ImgOps iqdb SauceNAO 2girls, NSFW, deeply affectionat (...).png, 2MiB, 1024x1024

Anonymous Mon 27 Nov 2023 09:46:40 No.7677058 Report

Quoted By:

This shit is incredable. Never managed to get decent yuri + face expressions so easily.

Anonymous

Anonymous Mon 27 Nov 2023 09:48:31 No.7677059 Report

Quoted By:

>>7677057
based
this is all about revenge, make some nai3 to kill localfags, it is backfiring

Anonymous

Anonymous Mon 27 Nov 2023 09:49:25 No.7677060 Report

Quoted By:

>>7677057
nice job

Anonymous

Anonymous Mon 27 Nov 2023 09:49:41 No.7677061 Report

Quoted By: >>7677064

>>7677057
you're proving models inferiority requiring "plugins and mods" each and every time you want something different.
but whatever, you do you, i'm just so fucking tired having a 30 gb concept lora folder and remembering which one i need to use for a specific prompt.

Anonymous

Anonymous Mon 27 Nov 2023 09:52:27 No.7677064 Report

Quoted By: >>7677069 >>7677071

>>7677061 (me)
by the way most concept loras are fucking cancer, you can't mix and match them, just because of loras nature.
they're bandaids, not a solution.

Anonymous

View Same Google ImgOps iqdb SauceNAO To Journey ever forwards.png, 3MiB, 1604x1664

Anonymous Mon 27 Nov 2023 09:52:52 No.7677065 Report

Quoted By:

>>7677052
We've come a long way. But at the end of our long road, we will find peace and Erika lewds.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1697199460584583.png, 994KiB, 768x1152

Anonymous Mon 27 Nov 2023 09:53:20 No.7677066 Report

Quoted By:

Here comes the mountain of cope.

Anonymous

Anonymous Mon 27 Nov 2023 09:57:17 No.7677069 Report

Quoted By: >>7677075

>>7677064
You can mix some of them and some of them can't. But yeah, dumb base model sucks ass. Combining character and concept/style loras will just end up in melting mess that you have to fiddle around with. Yes, style loras are essential although even they barely help against this vomitus inducing local 3d style. I can understand that some newfag that installed SD a month ago might still fucking love this fucking epic AOM look.

Anonymous

View Same Google ImgOps iqdb SauceNAO ComfyUI_00024_.png, 1MiB, 920x920

Anonymous Mon 27 Nov 2023 09:57:57 No.7677070 Report

Quoted By: >>7677525

>>7677009
"penetration through clothes" has less than 300 hits so even forcing it does nothing on local, but I've made plenty of these by accident. If you ask for say "sports bra" and do not specify "bottomless" it will give you spats or leggings even in a sex scene, about one in five times.

Anonymous

Anonymous Mon 27 Nov 2023 09:58:26 No.7677071 Report

Quoted By: >>7677073

>>7677064
nta, but loras or lora adjacent modules will likely still have a place if naiv3 is ever released. Even with SDXL, the base model does not have perfect weights for certain concepts.

Anonymous

Anonymous Mon 27 Nov 2023 10:00:08 No.7677073 Report

Quoted By: >>7677079

>>7677071
even if the base model had perfect weights for everything, that's stuck in time. if some new meme popped up like "praise the booty" or "mcmommy", someone would want it trained.
both are important, base model will always benefit from lora augmentation or further finetunes/merges.

Anonymous

Anonymous Mon 27 Nov 2023 10:00:43 No.7677075 Report

Quoted By:

>>7677069
I'm not gonna paypig but lack of aom style is so fucking refreshing

Anonymous

Anonymous Mon 27 Nov 2023 10:03:44 No.7677079 Report

Quoted By:

>>7677073
perfect model implies you can prompt anything into it, with natural language even. this is just about prompt comprehension.

Anonymous

View Same Google ImgOps iqdb SauceNAO motorcycle_0.gif, 1MiB, 2460x1640

Anonymous Mon 27 Nov 2023 10:04:08 No.7677080 Report

Quoted By:

Segmentation-Based Parametric Painting
https://arxiv.org/abs/2311.14271
>We introduce a novel image-to-painting method that facilitates the creation of large-scale, high-fidelity paintings with human-like quality and stylistic variation. To process large images and gain control over the painting process, we introduce a segmentation-based painting process and a dynamic attention map approach inspired by human painting strategies, allowing optimization of brush strokes to proceed in batches over different image regions, thereby capturing both large-scale structure and fine details, while also allowing stylistic control over detail. Our optimized batch processing and patch-based loss framework enable efficient handling of large canvases, ensuring our painted outputs are both aesthetically compelling and functionally superior as compared to previous methods, as confirmed by rigorous evaluations.
https://github.com/manuelladron/semantic_based_painting
neat

Anonymous

Anonymous Mon 27 Nov 2023 10:06:41 No.7677083 Report

Quoted By: >>7677087 >>7677088 >>7677089

some stupid questions I have had from the 2weeks I've been playing with local:
>Why is there no other booru or pixiv tag support?
>the likelyhood of NAI3 being leaked the same time as the original leak?

Anonymous

Anonymous Mon 27 Nov 2023 10:07:23 No.7677085 Report

Quoted By:

>>7677021
Lol how gay can you be

Anonymous

View Same Google ImgOps iqdb SauceNAO Untitled.png, 1MiB, 1687x763

Anonymous Mon 27 Nov 2023 10:10:03 No.7677086 Report

Quoted By:

Concept Sliders: LoRA Adaptors for Precise Control in Diffusion Models
https://arxiv.org/abs/2311.12092
>We present a method to create interpretable concept sliders that enable precise control over attributes in image generations from diffusion models. Our approach identifies a low-rank parameter direction corresponding to one concept while minimizing interference with other attributes. A slider is created using a small set of prompts or sample images; thus slider directions can be created for either textual or visual concepts. Concept Sliders are plug-and-play: they can be composed efficiently and continuously modulated, enabling precise control over image generation. In quantitative experiments comparing to previous editing techniques, our sliders exhibit stronger targeted edits with lower interference. We showcase sliders for weather, age, styles, and expressions, as well as slider compositions. We show how sliders can transfer latents from StyleGAN for intuitive editing of visual concepts for which textual description is difficult. We also find that our method can help address persistent quality issues in Stable Diffusion XL including repair of object deformations and fixing distorted hands.
https://sliders.baulab.info
https://github.com/rohitgandikota/sliders
https://sliders.baulab.info/weights/xl_sliders/
maybe old news and only for xl I guess so eh

Anonymous

Anonymous Mon 27 Nov 2023 10:10:43 No.7677087 Report

Quoted By:

>>7677083
>Why is there no other booru or pixiv tag support?
The base model was trained on images and tags from danbooru, so if you use tags from other sites like Pixiv it's less likely to recognise what they are
>the likelyhood of NAI3 being leaked the same time as the original leak?
Probably not very high. It (allegedly) took someone burning a day 0 exploit to leak the original one so you'd think most companies would try to tighten their security measures after such an event. Granted that the entire idea behind day 0s is circumventing those measures it still takes someone to not only find it but also use it

Anonymous

Anonymous Mon 27 Nov 2023 10:11:32 No.7677088 Report

Quoted By: >>7677089 >>7677094 >>7677101

>>7677083
>Why is there no other booru or pixiv tag support?
The model was trained with danbooru images and their tags only. Pixiv has japanese tags which just complicates things
>the likelyhood of NAI3 being leaked the same time as the original leak?
Near zero. They learned their lesson from last time and the new model is reason enough to have Area 51 level security

Anonymous

Anonymous Mon 27 Nov 2023 10:12:27 No.7677089 Report

Quoted By:

>>7677088
>>7677083
Pixiv tags are absolute fucking disaster anyway

Anonymous

View Same Google ImgOps iqdb SauceNAO __chima_monobeno_drawn_by_cura__ (...).jpg, 473KiB, 850x2613

Anonymous Mon 27 Nov 2023 10:13:17 No.7677090 Report

Quoted By: >>7677222

>>7676959
>>7676963
>>7677044

Yeah for them most part I was pleasantly surprised with how well it worked for a few VNs, sometimes if it didn't work, I could just describe their looks as well + year to narrow down the artist's style around that time and it'd nail it. Other times it just wouldn't work no matter what I did (Chima from Monobeno, Kuro from Sakura Moyu), makes me wish I was more proactive in uploading and tagging stuff on danbooru

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girls. 1boy, hyuuga hinat (...).png, 976KiB, 1024x1024

Anonymous Mon 27 Nov 2023 10:21:06 No.7677094 Report

Quoted By:

>>7677088
>Leak chance near zero
What a shame, honestly. That or that they won't ever publicly release the model in any way, despite them having greatly benefited from open source themselves and copied features right off controlnet.

Not to be depressive or go down a rabbit hole, but it extra sucks that corporations and governments will be the only people who have this kind of thing available to use.

Anonymous

Anonymous Mon 27 Nov 2023 10:25:42 No.7677101 Report

Quoted By: >>7677667 >>7677679

>>7677088
there was no 0day anon
nobody dumps their entire enterprises files on one production server, doesn't happen
the leak came from within and was pretty much intentional, as for why? who the fuck knows

Anonymous

Anonymous Mon 27 Nov 2023 10:28:31 No.7677104 Report

Quoted By: >>7677106

>>7677057
Which lora is that?

Anonymous

Anonymous Mon 27 Nov 2023 10:29:48 No.7677106 Report

Quoted By: >>7677118

>>7677104
https://civitai.com/models/69357/concept-spread-pussyfinger

Anonymous

Anonymous Mon 27 Nov 2023 10:32:28 No.7677112 Report

Quoted By:

>>7677057
so close, if only it looked a bit more like fabric

Anonymous

View Same Google ImgOps iqdb SauceNAO 2girls, deeply affectionate, {{{ (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 10:33:36 No.7677113 Report

Quoted By: >>7677116

Depressing imagining what people could do with this model if we had our hands on it.

Anonymous

Anonymous Mon 27 Nov 2023 10:34:53 No.7677115 Report

Quoted By: >>7677121

>>7677057
Catbox then

Anonymous

Anonymous Mon 27 Nov 2023 10:35:44 No.7677116 Report

Quoted By:

>>7677113
More of the stuff danbooru is already full of?

Anonymous

Anonymous Mon 27 Nov 2023 10:36:35 No.7677117 Report

Quoted By: >>7677120 >>7677121

>>7677057
I would pay hundred bucks a day to not ever see slop like this ever again

Anonymous

Anonymous Mon 27 Nov 2023 10:37:29 No.7677118 Report

Quoted By:

>>7677106
Thanks, there's way too many loras for spread pussy. Really cool, none of the civitai images used it this way.

Anonymous

Anonymous Mon 27 Nov 2023 10:38:24 No.7677119 Report

Quoted By:

>>7676962
What are y'all guys comfy Workflow and extensions? It seems powerful

Anonymous

Anonymous Mon 27 Nov 2023 10:39:35 No.7677120 Report

Quoted By: >>7677123

>>7677117
for the love of god please stop shitposting like this and let people enjoy themselves. you aren't making it better by throwing shit in the fan

Anonymous

Anonymous Mon 27 Nov 2023 10:39:43 No.7677121 Report

Quoted By: >>7677122 >>7677140

>>7677115
>original gen
https://files.catbox.moe/byjzj9.png
>15 seconds of selecting an object and changng hue in Photoshop
https://files.catbox.moe/ntx0v4.png
>5 seconds of inpaint at low denoise
https://files.catbox.moe/e1cf4r.png
>>7677117
Didn't your shilling guide tell you not to overuse the same words while trying to blend in?

Anonymous

Anonymous Mon 27 Nov 2023 10:41:07 No.7677122 Report

Quoted By: >>7677124

>>7677121
ah, so it was inpainted...

Anonymous

Anonymous Mon 27 Nov 2023 10:41:16 No.7677123 Report

Quoted By: >>7677125

>>7677120
Im sorry man I cant help myself
Its like this people are fucking blind

Anonymous

Anonymous Mon 27 Nov 2023 10:41:49 No.7677124 Report

Quoted By:

>>7677122
I said this from the beginning.

Anonymous

Anonymous Mon 27 Nov 2023 10:45:25 No.7677125 Report

Quoted By:

>>7677123
It's one ban evading guy. Ignore him and post pictures.

Anonymous

View Same Google ImgOps iqdb SauceNAO 2b185304-0d49-42b2-96c5-77a2044d (...).png, 2MiB, 1216x832

Anonymous Mon 27 Nov 2023 10:49:26 No.7677129 Report

Quoted By: >>7678041

Surprised it struggled so much with roxanne from isekai meikyuu, even with the description. Looking at danbooru I'm guessing some of her art got deleted or was somehow never uploaded to danbooru, weird considering how popular she was

Anonymous

Anonymous Mon 27 Nov 2023 10:50:09 No.7677130 Report

Quoted By: >>7677132 >>7677134

what's the best way to do a dark lighting/moonlight on local? anything I try results in a night scenery with lots of lamps even if I had them in negative prompts and the weights modified.

Anonymous

Anonymous Mon 27 Nov 2023 10:51:48 No.7677132 Report

Quoted By: >>7677134

>>7677130
you need a model/lora with noise offset (search for lowra

Anonymous

Anonymous Mon 27 Nov 2023 10:52:32 No.7677134 Report

Quoted By:

>>7677130
>>7585568
Maybe that would be of any help. Outdoors/landscape landscape sceneries won't have lamps genned
>>7677132
Yeah or lowra but I prefer the img2img

Anonymous

Anonymous Mon 27 Nov 2023 11:00:34 No.7677140 Report

Quoted By: >>7677143 >>7677145 >>7677718

>>7677121
What are your Workflow and extensions for comfyUI? I wanna give it a try cuz it seems powerful, i'm kinda tired of a1111

Anonymous

View Same Google ImgOps iqdb SauceNAO 536ba4ff-b4a8-4560-b6ae-afd7b7dc (...).png, 974KiB, 1216x832

Anonymous Mon 27 Nov 2023 11:02:32 No.7677142 Report

Quoted By: >>7677153

Time for a health checkup

Anonymous

View Same Google ImgOps iqdb SauceNAO 1700048411229086.png, 51KiB, 343x478

Anonymous Mon 27 Nov 2023 11:03:16 No.7677143 Report

Quoted By:

>>7677140
I don't use it, seems too complicated to me and it's missing crucial functionality added by plugins (tag autocomplete being by far the most important one I use daily, not just for tags but for LoRAs as well).

Anonymous

Anonymous Mon 27 Nov 2023 11:07:03 No.7677145 Report

Quoted By: >>7677146 >>7677150

>>7677140
>it seems powerful
it cannot achieve anything a1111 can't. It's the same shit just complicated as fuck - if you want to mimic it's regio prompting just use regional prompter, that's where most of the comfy "magic" comes from

Anonymous

Anonymous Mon 27 Nov 2023 11:10:36 No.7677146 Report

Quoted By: >>7677149

>>7677145
I can take the argument that comfy is more convenient for the developers
And with improved krita plugin they claim that the ui is not complete trash now. I am yet to try it though

Anonymous

Anonymous Mon 27 Nov 2023 11:15:25 No.7677149 Report

Quoted By: >>7677151 >>7677154 >>7677155

>>7677146
why would krita change anything there, it's still absolute ass because of the retarded nodes.
Everytime you want to change the prompt you're looking at over 30 minutes of work, thats why these retards only ever gen the same thing over and over again.
>comfy
I wouldn't trust anyone who tells you it's good - most of the shills are legit paid employees who wanted to make it big off owning *the* local UI. The other half is autistic retards that'd probably think linux is good aswell

Anonymous

Anonymous Mon 27 Nov 2023 11:16:50 No.7677150 Report

Quoted By:

>>7677145
Seems that way so far. It helped me understand how things work under the hood, stuff like having to assemble hiresfix from nodes. But I'm not sure if the knowledge gained was actually useful to me.

There are some things I haven't seen elsewhere, like having separate lora weights for model and prompts. Or doing a workflow with inpaint steps that doesn't need to deconde into a png image halfway through. And I'm sure there's more, but they're all niche use cases. If they weren't then somebody would implement them for a11 or the other tools.

Anonymous

Anonymous Mon 27 Nov 2023 11:18:35 No.7677151 Report

Quoted By: >>7677152

>>7677149
>don't have to touch nodes anymore
>it sucks because of the thing you don't have to interact with anymore still makes me mad
???

Anonymous

Anonymous Mon 27 Nov 2023 11:19:18 No.7677152 Report

Quoted By:

>>7677151
krita is an inpainting tool, no?

Anonymous

View Same Google ImgOps iqdb SauceNAO 74690-AI.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 11:19:33 No.7677153 Report

Quoted By: >>7680512

>>7677142
People actually get paid 6 figures to do this to horses

Anonymous

Anonymous Mon 27 Nov 2023 11:19:49 No.7677154 Report

Quoted By:

>>7677149
>Everytime you want to change the prompt you're looking at over 30 minutes of work,
Clueless

Anonymous

Anonymous Mon 27 Nov 2023 11:19:52 No.7677155 Report

Quoted By: >>7677163

>>7677149
I don't mind people choosing for themselves, if comfy could do anything at all that represented a killer feature I'd switch in a heartbeat. It's just a lot of work for basically no payoff.
The one thing I would shill for comfy is that it's objectively faster to gen than a1111, perhaps when new xl finetunes show up comfy will be *the* fastest option, until then it's just a more tedious way to gen and has fun nodes for people who want to factorio their way through porn.

Anonymous

Anonymous Mon 27 Nov 2023 11:25:26 No.7677157 Report

Quoted By:

>>7676936
>FURTHERMORE, characters are actually inputted during training as character:charactername and artists as artist:artistname.
This is not correct. Both with and without prefix were used.

Anonymous

Anonymous Mon 27 Nov 2023 11:30:26 No.7677158 Report

Quoted By: >>7677160 >>7677162

Is it possible to make animations with NovelAI?

Anonymous

Anonymous Mon 27 Nov 2023 11:30:50 No.7677160 Report

Quoted By:

>>7677158
nope

Anonymous

Anonymous Mon 27 Nov 2023 11:34:35 No.7677162 Report

Quoted By: >>7677169 >>7677170

>>7677158
Maybe if you made a 3d video and put all the frames through NAI with the relevant prompts and same seed. Don't know though, never messed around with animation

Anonymous

Anonymous Mon 27 Nov 2023 11:34:47 No.7677163 Report

Quoted By:

>>7677155
yes XL on a1111 objectively sucks ass. I have no idea why though it's still the case

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, aqua (kingdom heart (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 11:36:30 No.7677165 Report

Quoted By: >>7677199

>>7676989
SFW: https://files.catbox.moe/4419jl.png

Anonymous

Anonymous Mon 27 Nov 2023 11:41:48 No.7677169 Report

Quoted By: >>7677171 >>7677174

>>7677162
Quick rundown on seeds? Does it matter at all or is it just like upping the chance for a higher quality/more accurate image?

Anonymous

Anonymous Mon 27 Nov 2023 11:42:50 No.7677170 Report

Quoted By:

>>7677162
rotoshit always look like shit. it's basically comfyui only for videos now since all the competent vid devs moved onto it or is using it for their backends. maybe anon should make a node that hooks up to the api so people can animate at all.

Anonymous

Anonymous Mon 27 Nov 2023 11:43:02 No.7677171 Report

Quoted By:

>>7677169
google it

Anonymous

Anonymous Mon 27 Nov 2023 11:45:51 No.7677174 Report

Quoted By:

>>7677169
just go ask ani, I'm sure he can whip up a dozen ways to do it

Anonymous

View Same Google ImgOps iqdb SauceNAO 02580-4756428865.png, 1MiB, 1024x1280

Anonymous Mon 27 Nov 2023 11:57:16 No.7677180 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 12:17:48 No.7677195 Report

Quoted By: >>7677197

cant make kirie from girls bravo

Anonymous

Anonymous Mon 27 Nov 2023 12:18:19 No.7677197 Report

Quoted By: >>7677200 >>7677218 >>7677220

>>7677195
>59 images on Danbooru

Anonymous

View Same Google ImgOps iqdb SauceNAO aqasd.png, 823KiB, 768x1152

Anonymous Mon 27 Nov 2023 12:19:40 No.7677199 Report

Quoted By: >>7677201 >>7677206 >>7677219

>>7677165
That's not bad, it even gets her weird mullet right.
Pic related was probably the best result I got with some lora I cooked

Is Nai v3 worth the price?

Anonymous

Anonymous Mon 27 Nov 2023 12:21:50 No.7677200 Report

Quoted By: >>7677218 >>7677221 >>7677223 >>7677277

>>7677197
I mean did they actually ONLY use danbooru images in their dataset?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00630-4274173128.png, 1MiB, 1280x1024

Anonymous Mon 27 Nov 2023 12:22:28 No.7677201 Report

Quoted By: >>7677211

>>7677199
it is fun to play around with, any claims it destroys local are unfounded

Anonymous

Anonymous Mon 27 Nov 2023 12:24:01 No.7677204 Report

Quoted By:

>>7676979
>the furries the furries the furries
the furfaggots and localoids in general have been playing catch up
also the furfags have been using jewgle tpu gibs, they're poorfags
the rich ones still cummission artoids as usual

Anonymous

Anonymous Mon 27 Nov 2023 12:24:59 No.7677206 Report

Quoted By:

>>7677199
if you like the NAI pics you see it's worth it, if not really then don't bother
for me it was absolutely worth it because I never was able to fap to NAI1/local hentai, it always looked extremely off to me but if you like it, why waste money

Anonymous

Anonymous Mon 27 Nov 2023 12:25:45 No.7677207 Report

Quoted By: >>7677216

>>7676982
NAI V3 can do 2 characters guaranteed and 3 with some very quick gacha
4 is too hit and miss

Anonymous

Anonymous Mon 27 Nov 2023 12:27:15 No.7677211 Report

Quoted By: >>7677233

>>7677201
>any claims it destroys local are unfounded
yeah completely unfounded save for all the images in the previous threads <- plural

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, 1boy, {{matilda fiship}}, (...).png, 2MiB, 1536x1024

Anonymous Mon 27 Nov 2023 12:30:07 No.7677215 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 12:30:34 No.7677216 Report

Quoted By: >>7677229

>>7677207
Just git gud and learn to region proompt.

Anonymous

Anonymous Mon 27 Nov 2023 12:30:55 No.7677217 Report

Quoted By: >>7677271 >>7677289

Do anyone has a example of a very well tagged dataset?

Anonymous

Anonymous Mon 27 Nov 2023 12:31:02 No.7677218 Report

Quoted By: >>7677277

>>7677197
>>7677200
Wait, really? Is there a way to pull from other sources and not just danbooru?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684917833552488.png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 12:31:05 No.7677219 Report

Quoted By:

>>7677199
it lets me do lovey dovey images and turns girls into gyarus. Incredibly comfy and fun to use with a wide artist range so yeah I'd say it is.
Tempted to unsub next month just to stop torturing my cock

Anonymous

Anonymous Mon 27 Nov 2023 12:32:37 No.7677220 Report

Quoted By:

>>7677197
that doesnt really tell us anything since current NAI might be able to do it, probably can in fact

Anonymous

Anonymous Mon 27 Nov 2023 12:32:59 No.7677221 Report

Quoted By:

>>7677200
Yes.

Anonymous

Anonymous Mon 27 Nov 2023 12:33:01 No.7677222 Report

Quoted By: >>7677235

>>7677090
this one could have fooled me, and i've been staring at image generators for months.

Anonymous

Anonymous Mon 27 Nov 2023 12:34:07 No.7677223 Report

Quoted By: >>7677254 >>7677277

>>7677200
yes and they went balls deep with it, they fucking fried it on danbooru. which is a good thing if you want to coom but not so much otherwise.

Anonymous

Anonymous Mon 27 Nov 2023 12:38:40 No.7677229 Report

Quoted By: >>7677232

>>7677216
Silence localoid, you need regional prompting to even do 2 characters consistently.

Anonymous

Anonymous Mon 27 Nov 2023 12:39:36 No.7677232 Report

Quoted By: >>7677238 >>7677259

>>7677229
uhhh 2 characters do get their characteristics confused on nai too, although it's not as messy as local

Anonymous

View Same Google ImgOps iqdb SauceNAO 01252-4402456309.png, 1MiB, 1280x1024

Anonymous Mon 27 Nov 2023 12:41:36 No.7677233 Report

Quoted By: >>7677239

>>7677211
>images in the previous threads
for the uninitiated

Anonymous

Anonymous Mon 27 Nov 2023 12:46:25 No.7677235 Report

Quoted By:

>>7677222
ah sorry, that wasn't a gen, that was an example of a character i couldn't gen in case someone can prove me wrong and gen her LOL

Anonymous

Anonymous Mon 27 Nov 2023 12:48:16 No.7677238 Report

Quoted By:

>>7677232
Yeah I did a monobeno maitetsu one on another thread and it took like 50 gens to not get the hairstyles and eye colors mixed up. NAI3 could definitely benefit from some controlnets and regional prompting

Anonymous

Anonymous Mon 27 Nov 2023 12:48:32 No.7677239 Report

Quoted By: >>7677745

>>7677233
do it again with her head not upright

Anonymous

View Same Google ImgOps iqdb SauceNAO 00187-1063856542.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 13:00:08 No.7677253 Report

Quoted By: >>7678379

>>7676552
>>7676726
catbox?

Anonymous

Anonymous Mon 27 Nov 2023 13:01:37 No.7677254 Report

Quoted By: >>7677255 >>7677284

>>7677223
So all we have to do is mass upload everything that's not yet on Danbooru to Danbooru? And then it'll be equal to Dalle3 in terms of scope?

Anonymous

Anonymous Mon 27 Nov 2023 13:03:05 No.7677255 Report

Quoted By:

>>7677254
that's some troll physics stuff, I like this approach

Anonymous

View Same Google ImgOps iqdb SauceNAO 2girls, imaizumi kagerou, wakasa (...).jpg, 764KiB, 1216x832

Anonymous Mon 27 Nov 2023 13:07:34 No.7677259 Report

Quoted By:

>>7677232
skill issue

Anonymous

Anonymous Mon 27 Nov 2023 13:08:44 No.7677261 Report

Quoted By: >>7677263 >>7677264

Why are samplers so fucking different on NAI
the differences are not as strong on local (for same seed)

Anonymous

Anonymous Mon 27 Nov 2023 13:09:55 No.7677263 Report

Quoted By: >>7677265

>>7677261
I'd prefer a handful of vastly different samplers over the many samey samplers on local.

Anonymous

Anonymous Mon 27 Nov 2023 13:10:01 No.7677264 Report

Quoted By: >>7677265

>>7677261
What are you talking about? They matter even more on V3

Anonymous

Anonymous Mon 27 Nov 2023 13:11:40 No.7677265 Report

Quoted By:

>>7677264
>They matter even more on V3
well yeah that's what I meant?
>>7677263
probably yes but their behavior is inconsistent. sometimes euler looks better sometimes 2sa
they all also seem to work better at 3-5 cfg, higher kinda fries

Anonymous

Anonymous Mon 27 Nov 2023 13:14:41 No.7677267 Report

Quoted By: >>7677270 >>7677272

Is it pulling from Danbooru in real time, meaning I can upload a gallery from an artist and I can prompt it? Or does the program need an update and it needs to internalize Danbooru for each version

Anonymous

Anonymous Mon 27 Nov 2023 13:16:58 No.7677270 Report

Quoted By: >>7677283

>>7677267
It's based on Danbooru some time around in the summer of this year. Don't expect it to get updated until V4

Anonymous

Anonymous Mon 27 Nov 2023 13:17:48 No.7677271 Report

Quoted By:

>>7677217
I guess no one knows what they're actually doing.

Anonymous

Anonymous Mon 27 Nov 2023 13:18:04 No.7677272 Report

Quoted By:

>>7677267
>Is it pulling from Danbooru in real time, meaning I can upload a gallery from an artist and I can prompt it?
no that's not even remotely how it works. it's a static model, on local it's a 2-6gb file. with nai it's a snapshot of the time when they trained it
if they want to update it they need to continue training or retrain the model, updating the file this way

Anonymous

View Same Google ImgOps iqdb SauceNAO 94002d9e-ac20-4b90-8aee-73b9c203 (...).png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 13:19:51 No.7677273 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 13:21:45 No.7677275 Report

Quoted By:

>>7676592
they're both SD

Anonymous

Anonymous Mon 27 Nov 2023 13:25:10 No.7677277 Report

Quoted By: >>7677279

>>7677218
>>7677223
>>7677200
dude, what the should have done is invest as much time and compute power building a tagger as they did nai3 so they could expand their dataset even further beyond.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, aqua (kingdom heart (...).png, 848KiB, 512x1408

Anonymous Mon 27 Nov 2023 13:26:39 No.7677278 Report

Quoted By: >>7677417 >>7677536

She has a great sense of balance.

Anonymous

Anonymous Mon 27 Nov 2023 13:26:49 No.7677279 Report

Quoted By:

>>7677277
maybe just maybe the real issue is booru tags being just a list instead of branching

Anonymous

Anonymous Mon 27 Nov 2023 13:27:44 No.7677281 Report

Quoted By:

>>7676936
why are you acting like this information is some secret source when it's literally just reading their public documentation and the tooltips

http://docs.novelai.net/image/qualitytags.html

Anonymous

Anonymous Mon 27 Nov 2023 13:27:56 No.7677283 Report

Quoted By: >>7677285

>>7677270
It is kind of hilarious, but you can kind of pinpoint the date if you pretty much test out whatever major girl launched in a mobile game that month

Anonymous

Anonymous Mon 27 Nov 2023 13:28:53 No.7677284 Report

Quoted By:

>>7677254
Looks like it. Make sure your favorite characters are well tagged and have 150+ images and they'll probably show up in future models.

Anonymous

Anonymous Mon 27 Nov 2023 13:29:17 No.7677285 Report

Quoted By:

>>7677283
Some anons were doing just that a week or two ago

Anonymous

Anonymous Mon 27 Nov 2023 13:30:07 No.7677286 Report

Quoted By: >>7677287 >>7677288 >>7677290 >>7677293 >>7677295

Is it better that Danbooru is missing a lot of stuff? Less low quality pictures to dilute the good stuff? Or should we "complete" Danbooru

Anonymous

Anonymous Mon 27 Nov 2023 13:32:06 No.7677287 Report

Quoted By:

>>7677286
I don't know if you gain more from not having some bad pictures over not having cg rips

Anonymous

Anonymous Mon 27 Nov 2023 13:32:25 No.7677288 Report

Quoted By: >>7677296 >>7678041

>>7677286
you can't "complete" it, artists will just issue takedowns

Anonymous

Anonymous Mon 27 Nov 2023 13:32:48 No.7677289 Report

Quoted By: >>7677552

>>7677217
yeah, mine

Anonymous

Anonymous Mon 27 Nov 2023 13:33:01 No.7677290 Report

Quoted By:

>>7677286
I think sankaku is the more "complete" danbooru but it's probably a pain in the ass to scrape it
and it really is full of shit

Anonymous

Anonymous Mon 27 Nov 2023 13:33:53 No.7677293 Report

Quoted By: >>7678041

>>7677286
A ton of content isn't allowed, and many artists have been blacklisted or whatever. Currently, the only way to access loli is to upload 100's of images, so there really aren't too many holes left to fill that can be filled.

Anonymous

Anonymous Mon 27 Nov 2023 13:34:29 No.7677295 Report

Quoted By: >>7677297

>>7677286
Low quality pictures aren’t the issue, low quality tagging is.

Anonymous

Anonymous Mon 27 Nov 2023 13:34:36 No.7677296 Report

Quoted By: >>7677307

>>7677288
Okay, but you said that it's not a "real time" pulling. So all we gotta do is upload the paywalled stuff when the "collecting phase" is happening so even if they do get taken down, the rest of the art is caught in the "snapshot"

Anonymous

Anonymous Mon 27 Nov 2023 13:35:46 No.7677297 Report

Quoted By:

>>7677295
Both are the problem. Anyone who's trained a lora and had their dataset poisoned by one (1) image could tell you that.

Anonymous

Anonymous Mon 27 Nov 2023 13:36:28 No.7677299 Report

Quoted By: >>7677302

>this is the kind of retard that actually pays for NAI
bruh

Anonymous

Anonymous Mon 27 Nov 2023 13:37:20 No.7677301 Report

Quoted By: >>7677305 >>7677308 >>7677309

Wouldn't it be worth it to invest some time in creating a good autotagger so you wouldn't have to deal with boorus? Could just scrape twitter or some other site without paywall. It's what I do with a lot of low image artists when creating loras

Anonymous

Anonymous Mon 27 Nov 2023 13:37:21 No.7677302 Report

Quoted By:

>>7677299
We got it, you love AOM3. Fuck off already.

Anonymous

Anonymous Mon 27 Nov 2023 13:39:15 No.7677305 Report

Quoted By:

>>7677301
i wouldnt know where to start creating an autotagger

Anonymous

Anonymous Mon 27 Nov 2023 13:40:29 No.7677306 Report

Quoted By:

The problem isn't just the shit tagging, it's not having branching tags

Anonymous

Anonymous Mon 27 Nov 2023 13:41:27 No.7677307 Report

Quoted By: >>7678041

>>7677296
>Okay, but you said that it's not a "real time" pulling
I didn't say that another anon did
>So all we gotta do is upload the paywalled stuff when the "collecting phase" is happening
maybe, but for the artists who have already got their shit blocked that isn't gonna work and danbooru has a shit load of artists who already request that their stuff be removed from when ai art first took off

Anonymous

Anonymous Mon 27 Nov 2023 13:41:31 No.7677308 Report

Quoted By:

>>7677301
auto taggers are ... guess what trained on

Anonymous

Anonymous Mon 27 Nov 2023 13:41:36 No.7677309 Report

Quoted By:

>>7677301
well multimodal llms are a thing. but i've yet to see a convincing example of open source models being able to tag images correctly.
but yeah, i presume it's what dalle3 used for training for natural language comprehension.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{artist_ fizrotart}}}, shoebil (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 13:47:58 No.7677313 Report

Quoted By: >>7677317

Why aren't bird girls real? I just want to take a relaxing stroll on the beach with my wife before impregnating her again (again)

Anonymous

Anonymous Mon 27 Nov 2023 13:50:17 No.7677315 Report

Quoted By: >>7677320 >>7677324 >>7677325 >>7677332

I'm taking a break from prooompting today
I've spent literally too much of the past 3-4 days erect and I can feel my lower body starting to ache
Today will be a day of recuperation for my cock

Anonymous

Anonymous Mon 27 Nov 2023 13:50:36 No.7677317 Report

Quoted By: >>7677322

>>7677313
The pidgeons in the park are free, you can just take them.

Anonymous

Anonymous Mon 27 Nov 2023 13:51:08 No.7677320 Report

Quoted By:

>>7677315
same but then i woke up and proompted anyway haha

Anonymous

Anonymous Mon 27 Nov 2023 13:51:40 No.7677321 Report

Quoted By:

>>7676992
Whenever I enhance at 4/5 it just deep fries the image and the lines/art style becomes even more blurred or the lines become way too strong
Particularly with stuff like mimonel or CSR

Anonymous

Anonymous Mon 27 Nov 2023 13:53:37 No.7677322 Report

Quoted By:

>>7677317
No. ಠ_ಠ

Anonymous

Anonymous Mon 27 Nov 2023 13:55:23 No.7677324 Report

Quoted By:

>>7677315
It feels like I broke my brain with non-stop pressing the button

Anonymous

Anonymous Mon 27 Nov 2023 13:56:04 No.7677325 Report

Quoted By:

>>7677315
head empty
balls empty

Anonymous

Anonymous Mon 27 Nov 2023 13:56:42 No.7677328 Report

Quoted By: >>7677329

It's a shame that Danbooru doesn't have enough.....you know.....

Anonymous

Anonymous Mon 27 Nov 2023 13:57:30 No.7677329 Report

Quoted By: >>7677331

>>7677328
we don't know, enough what? loli?

Anonymous

Anonymous Mon 27 Nov 2023 13:58:04 No.7677331 Report

Quoted By: >>7677339

>>7677329
It has that though.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, {{artist_jinroku}}, black (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 13:58:34 No.7677332 Report

Quoted By:

>>7677315
My dick needs a break, but my curiosity continues, so I'm just going to keep testing artists out because sometime the results are interesting considering the low amount of danbooru images.

Anonymous

Anonymous Mon 27 Nov 2023 13:59:09 No.7677335 Report

Quoted By: >>7677336 >>7677337 >>7677338 >>7677340

What's the trick to reducing blurryness in NAI?

Anonymous

Anonymous Mon 27 Nov 2023 14:00:20 No.7677336 Report

Quoted By: >>7677347

>>7677335
You can try the enhance feature without upscaling the picture, but it's not 100% reliable.

Anonymous

Anonymous Mon 27 Nov 2023 14:00:31 No.7677337 Report

Quoted By: >>7677347

>>7677335
Using artist tags? The characters aren't really blurry though and the backgrounds are always gonna be shit either way

Anonymous

Anonymous Mon 27 Nov 2023 14:01:11 No.7677338 Report

Quoted By:

>>7677335
"glasses" in face section

Anonymous

Anonymous Mon 27 Nov 2023 14:01:19 No.7677339 Report

Quoted By: >>7677344

>>7677331
Then what anon?

Anonymous

Anonymous Mon 27 Nov 2023 14:01:55 No.7677340 Report

Quoted By: >>7677347

>>7677335
if it's backgrounds you can't, they don't give a shit about those apparently

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{star guardian ahri}}}, {{{lea (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 14:01:59 No.7677341 Report

Quoted By: >>7677348

My wife

Anonymous

Anonymous Mon 27 Nov 2023 14:02:13 No.7677342 Report

Quoted By: >>7677372 >>7677506

Has anyone managed to get a Doggystyle POV shot looking down at the ass/penetration from above

Anonymous

Anonymous Mon 27 Nov 2023 14:02:40 No.7677344 Report

Quoted By:

>>7677339
I don't know, but I'd like anon to tell us.

Anonymous

Anonymous Mon 27 Nov 2023 14:04:40 No.7677347 Report

Quoted By: >>7677358 >>7677359 >>7678093

>>7677340
>>7677336
>>7677337
Maybe blurry is the wrong word, I guess it's more like sloppy brushwork? I've tried Mimonel and nyantcha before and it's almost never as clean as >>7676923 or >>7676973
Maybe I just need to roll the gacha more and mess around with enhance though

Anonymous

Anonymous Mon 27 Nov 2023 14:04:41 No.7677348 Report

Quoted By: >>7677351 >>7677352 >>7677355

>>7677341
Seriously? THAT costume? The generic blonde maho shoujo one?

Anonymous

Anonymous Mon 27 Nov 2023 14:06:15 No.7677350 Report

Quoted By:

>>7676776
there is noise scanning tech for a reason

>>7676769
it means he doesn't know how to use "cameltoe, spread pussy under clothes, pussy juice stain"

Anonymous

View Same Google ImgOps iqdb SauceNAO 00754-2837770541.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 14:06:49 No.7677351 Report

Quoted By: >>7677353 >>7677354

>>7677348
sg ahri is hot, you need a better lora though other ahri poster

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{star guardian ahri}}}, {{{lea (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 14:07:16 No.7677352 Report

Quoted By: >>7677353

>>7677348
Kys, star guardian ahri is top tier.

Anonymous

Anonymous Mon 27 Nov 2023 14:08:08 No.7677353 Report

Quoted By:

>>7677351
>>7677352
I prefer the old default skin.

Anonymous

Anonymous Mon 27 Nov 2023 14:08:21 No.7677354 Report

Quoted By: >>7677356

>>7677351
Was I supposed to think this was the Twin Towers attack from the thumbnail

Anonymous

Anonymous Mon 27 Nov 2023 14:08:27 No.7677355 Report

Quoted By:

>>7677348
its a nice costume

Anonymous

View Same Google ImgOps iqdb SauceNAO 911compareahri.jpg, 1MiB, 2048x1530

Anonymous Mon 27 Nov 2023 14:09:19 No.7677356 Report

Quoted By:

>>7677354
lol

Anonymous

Anonymous Mon 27 Nov 2023 14:09:23 No.7677357 Report

Quoted By: >>7677361

>>7677057
based btfo'd retards

Anonymous

Anonymous Mon 27 Nov 2023 14:09:51 No.7677358 Report

Quoted By: >>7677359

>>7677347
>I guess it's more like sloppy brushwork?
I know exactly what you mean and unfortunately it seems like it's going to be one of the defining characteristics of NAIv3 images. Best advice I can give is to play around with artist tags since they greatly influence what the image ends up looking like. Unfortunately this technically also pidgeonholes you into only using certain artists/artstyles

Anonymous

Anonymous Mon 27 Nov 2023 14:11:04 No.7677359 Report

Quoted By:

>>7677358
>>7677347
default quality tag "very aesthetic" strongly wrangles the output, take that shit out and append the other quality tags

Anonymous

View Same Google ImgOps iqdb SauceNAO 00236-[TFT]-87830505.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 14:11:36 No.7677360 Report

Quoted By: >>7677362 >>7677370 >>7677377 >>7677394

is there any reason to care about nai v3 yet if you have workflows with controlnets and loras set up already, I have not been impressed with most "amazing" examples or thought the same thing couldn't be done better with lora/controlnet

Anonymous

Anonymous Mon 27 Nov 2023 14:11:50 No.7677361 Report

Quoted By: >>7677365

>>7677357
this... I can no longer fap to anything that doesn't look like 3d render

Anonymous

Anonymous Mon 27 Nov 2023 14:12:44 No.7677362 Report

Quoted By: >>7677366

>>7677360
yeah if you only care about 1girl, smile, $$seasonalwaifucostume you don't need it at all

Anonymous

Anonymous Mon 27 Nov 2023 14:12:49 No.7677363 Report

Quoted By: >>7677369 >>7677387

Any way to sort away an artists old art from NAI? Can't really get Aoi Nagisa's current style

Anonymous

Anonymous Mon 27 Nov 2023 14:13:42 No.7677365 Report

Quoted By:

>>7677361
yeah I think that's you anon.

Anonymous

View Same Google ImgOps iqdb SauceNAO lappcomputer.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 14:14:10 No.7677366 Report

Quoted By: >>7677368 >>7677384 >>7677394

>>7677362
way to miss the point
Make your best lappland using a computer using only nai v3 then

Anonymous

Anonymous Mon 27 Nov 2023 14:14:49 No.7677367 Report

Quoted By:

There's definitely something wrong with danbooru. I looked up Tsunade, as a well-known character from the most normie series and like 20 years to accumulate art, she has barely 1K. Less than many Blue Archive whores.

Anonymous

Anonymous Mon 27 Nov 2023 14:15:13 No.7677368 Report

Quoted By:

>>7677366
Who?

Anonymous

Anonymous Mon 27 Nov 2023 14:15:23 No.7677369 Report

Quoted By:

>>7677363
"year 2023" or whatever

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{artist_ fizrotart}}}, keqing (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 14:16:08 No.7677370 Report

Quoted By: >>7677371 >>7677373 >>7677374 >>7677394 >>7677578

>>7677360
5 seconds with the laziest prompt I could come up with, first gen so I didn't even cherry-pick

Anonymous

Anonymous Mon 27 Nov 2023 14:16:50 No.7677371 Report

Quoted By: >>7677401

>>7677370
that looks pretty fried bro

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, princess peach, nin (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 14:17:06 No.7677372 Report

Quoted By: >>7677375 >>7677417 >>7677523

>>7677342

Anonymous

Anonymous Mon 27 Nov 2023 14:17:09 No.7677373 Report

Quoted By: >>7677401

>>7677370
but that is vastly worse

Anonymous

Anonymous Mon 27 Nov 2023 14:17:52 No.7677374 Report

Quoted By: >>7677386 >>7677401

>>7677370
NAIshills btfoing themselves in real time lmao

Anonymous

Anonymous Mon 27 Nov 2023 14:18:02 No.7677375 Report

Quoted By:

>>7677372
WAHOO

Anonymous

Anonymous Mon 27 Nov 2023 14:20:09 No.7677376 Report

Quoted By: >>7677439

request to prompt 'side view, upper body, {{char}} holding filled condom on her tongue, looking at viewer' on naiv3

Anonymous

Anonymous Mon 27 Nov 2023 14:21:08 No.7677377 Report

Quoted By: >>7677378 >>7677379 >>7677382

>>7677360
god fuck I can't look at 1.5 anymore
I now understand why ic was saying that all AI looks literally fucking same

Anonymous

View Same Google ImgOps iqdb SauceNAO mangayae.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 14:22:08 No.7677378 Report

Quoted By:

>>7677377
you can easily change styles

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, artist_nikorashi-ka, blac (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 14:22:16 No.7677379 Report

Quoted By:

>>7677377
AOM was a mistake.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00344-[TFT]-8783426.png, 2MiB, 1536x1536

Anonymous Mon 27 Nov 2023 14:23:23 No.7677380 Report

Quoted By: >>7677381 >>7678086

can someone try this non standard pose on naiv3 then? I just generated this first try with a lora no in-painting or cherry picking

female pov, masturbating

Anonymous

Anonymous Mon 27 Nov 2023 14:23:57 No.7677381 Report

Quoted By:

>>7677380
t.roon

Anonymous

Anonymous Mon 27 Nov 2023 14:24:49 No.7677382 Report

Quoted By:

>>7677377
been feeling this way for few months already, can tell 1.5 from a thumbnail 99% of time no matter how many cope loras dude applies

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{anila (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 14:25:15 No.7677383 Report

Quoted By: >>7677581 >>7677584 >>7677633

>>7677057
For me, it's just that not only is novelai higher quality, but it's the ease of use. Using local is like
>hit generate
>Wait long ass time
>image is shit
>regen
>wait
>shit
>regen
>it's alright
>inpaint and all that other shit
>after 30 minutes or more get a decent image
Plus most of the time the quality will be bad, getting good generations is the exception. Plus you need to wrangle local little every step of the fucking way. Novelai I just proompt, wait like 10 seconds, and if I don't like the image I just regen, it's all quick, AND most of the time the results are good, the bad ones are the minority. Like you will NEVER see something like this come from local

Anonymous

View Same Google ImgOps iqdb SauceNAO very aesthetic, nsfw, 1girl, lap (...).png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 14:25:27 No.7677384 Report

Quoted By: >>7677386 >>7677394

>>7677366
>your best
I don't even know her, so here's a quickie instead.

Anonymous

Anonymous Mon 27 Nov 2023 14:25:54 No.7677385 Report

Quoted By: >>7677388 >>7677391

>>hit generate
>>Wait long ass time
stopped reading right there
get an actual PC

Anonymous

Anonymous Mon 27 Nov 2023 14:26:19 No.7677386 Report

Quoted By:

>>7677384
>>7677374

Anonymous

Anonymous Mon 27 Nov 2023 14:26:56 No.7677387 Report

Quoted By: >>7677475

>>7677363
depends on the artist, if they have a lot of 2023 images then year 2023 may work, otherwise you're gonna have to settle for 2022 or 2021. in my case with shida kazuhiro, his older works poisoned the part of his artstyle i liked, luckily that artstyle was all contained within the tag for the visual nove that has that artstyle in question, 'wiz anniversary'
putting that in negative got the right artstyle out

Anonymous

Anonymous Mon 27 Nov 2023 14:27:02 No.7677388 Report

Quoted By: >>7677389 >>7677393

>>7677385
you don't have 4090 you have no room to talk either

Anonymous

Anonymous Mon 27 Nov 2023 14:27:34 No.7677389 Report

Quoted By: >>7677396 >>7677401

>>7677388
works fast enough on my 3080 nigger

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{anila (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 14:27:54 No.7677391 Report

Quoted By: >>7677395 >>7677398 >>7677633

>>7677385
I have a 3080 which does me good for anything gaming wise. So you're telling me I should spend a thousand or more on a new card to generate on a worse model instead of just spending 25$?

Anonymous

Anonymous Mon 27 Nov 2023 14:28:35 No.7677393 Report

Quoted By:

>>7677388
So it was nvidia shills all along

Anonymous

Anonymous Mon 27 Nov 2023 14:28:49 No.7677394 Report

Quoted By:

>>7677370
>>7677360

>>7677366
>>7677384

we proved it, naiv3 is absolutely as good as any local models

Anonymous

Anonymous Mon 27 Nov 2023 14:28:52 No.7677395 Report

Quoted By: >>7677403

>>7677391
No, you should stop being a needy brat and thinking a 3080 is in any way, shape or form "slow" at imagen.

Anonymous

Anonymous Mon 27 Nov 2023 14:28:54 No.7677396 Report

Quoted By:

>>7677389
>he can only afford a 3080
local is just not sending it's best I see

Anonymous

Anonymous Mon 27 Nov 2023 14:29:34 No.7677397 Report

Quoted By: >>7677401

4090 is a fucking joke, that card alone does cost minimum $25 in power bill per month just existing in your pc

Anonymous

Anonymous Mon 27 Nov 2023 14:29:44 No.7677398 Report

Quoted By: >>7677400

>>7677391
Yes. You see, spending money on a monopolist is a good thing. You aren't a poorfag, are you?

Anonymous

Anonymous Mon 27 Nov 2023 14:30:09 No.7677399 Report

Quoted By:

>>7676672
Japan does not do computers, so none of them had graphic cards lol.

This is people trying it out who couldn't do local at all

Anonymous

Anonymous Mon 27 Nov 2023 14:31:07 No.7677400 Report

Quoted By:

>>7677398
>Yes. You see, spending money on a monopolist is a good thing.
Very true. Donate to NVIDIA. Fuck small companies like NAI though. They're too greedy.

Anonymous

View Same Google ImgOps iqdb SauceNAO 278-nsfw,1girl, 1boy, cowgirl po (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 14:31:14 No.7677401 Report

Quoted By: >>7677405 >>7677406

>>7677371
>>7677373
>>7677374
Now go see what your first result is in any 1.5 local model, if you're coping so hard you can't grasp what what it means to get that far with zero effort, you have brain problems.

>>7677389
>works fast enough on my 3080 nigger
Embarassing, I have a 4090 and I still consider most gens to take a long ass time.

>>7677397
The 4090 is the most power efficient card there is. Literally. It consumes a fraction of the amount of power of the 3080 to generate the same image.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{anila (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 14:31:27 No.7677403 Report

Quoted By: >>7677405 >>7677729

>>7677395
>You can't call it slow because.........I don't think it is
I'll tell you one thing, novelai is for sure FASTER

Anonymous

Anonymous Mon 27 Nov 2023 14:33:02 No.7677405 Report

Quoted By:

>>7677401
>>7677403
holy gluttons

Anonymous

Anonymous Mon 27 Nov 2023 14:33:08 No.7677406 Report

Quoted By: >>7677413

>>7677401
>The 4090 is the most power efficient card there is. Literally. It consumes a fraction of the amount of power of the 3080 to generate the same image.
that's not the point and you know it. people just batch 8+ on it and click the gen button faster

or go much more fancy and do video

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 14:39:16 No.7677409 Report

Quoted By: >>7677410 >>7677412 >>7677418

I like novelai

Anonymous

Anonymous Mon 27 Nov 2023 14:40:41 No.7677410 Report

Quoted By: >>7677414 >>7677417

>>7677409
it's nice for copyright characters and futa on female. everything else i gen is sorely lacking without loras

Anonymous

Anonymous Mon 27 Nov 2023 14:41:51 No.7677412 Report

Quoted By:

>>7677409
NOOOOOOOOOOOOOOO
SHILL
You can't like things I don't like!

Anonymous

View Same Google ImgOps iqdb SauceNAO 362-nsfw1girl 1boy cowgirl posit (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 14:44:19 No.7677413 Report

Quoted By:

>>7677406
A 4090 underclocked/undervolted with a 5-10% loss in it/s consumes significantly less power than the 3080 underclocked with the same performance loss.

I have a 3080ti and a 4090 in my machine, so this is a direct comparison with actual hardware. The 3080 is completely unusable at the reduced power levels the 4090 runs efficiently at.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 14:44:46 No.7677414 Report

Quoted By: >>7677416

>>7677410
I have had legit no issues genning things I want, it's better imo because you don't need to download and swap between loras for basic shit

Anonymous

Anonymous Mon 27 Nov 2023 14:45:58 No.7677416 Report

Quoted By: >>7677429

>>7677414
i don't want basic shit and any 1girl pinups i like my backgrounds, simple white background just doesnt do it for me, if i wanted that i'd just open a booru

Anonymous

Anonymous Mon 27 Nov 2023 14:46:15 No.7677417 Report

Quoted By: >>7677426

>>7677410
How long would a scene like >>7677372 take to make? Or >>7677278?

Anonymous

Anonymous Mon 27 Nov 2023 14:46:21 No.7677418 Report

Quoted By:

>>7677409
I can say that V3 has definitely reignited my joy of prompting. It's nice to just be able to slowly build an idea up from start to finish just using words instead of having to fiddle with other things.
My biggest issue right now is the 75 token split where Token 76 and 77 literally do not work.

Anonymous

Anonymous Mon 27 Nov 2023 14:46:47 No.7677419 Report

Quoted By: >>7677423 >>7677429

>swap between loras for basic shit
another shill outed himself lol
basic shit is exactly what you DON'T need loras for, you need them for specific concepts or styles

Anonymous

Anonymous Mon 27 Nov 2023 14:47:43 No.7677420 Report

Quoted By: >>7677424 >>7677434 >>7677443 >>7677585

all these FAGGOTS with "4090s" and "3090s" who only shit on /hdg/ instead of finetuning a new local naiv4.
even furfags are laughing at you.

Anonymous

Anonymous Mon 27 Nov 2023 14:48:55 No.7677423 Report

Quoted By:

>>7677419
what? you totally need some lora for basic concept that should be done by the model take your meds please.

Anonymous

Anonymous Mon 27 Nov 2023 14:49:10 No.7677424 Report

Quoted By: >>7677427 >>7677453 >>7677455

>>7677420
I have an A6000, but I refuse to tune a model for you because it's funny.

Anonymous

Anonymous Mon 27 Nov 2023 14:49:23 No.7677426 Report

Quoted By: >>7677430 >>7677436 >>7677443 >>7677457

>>7677417
thats a basic pov shot which local can do fine, and for the second you could controlnet it but honestly the only thing naiv3 did right in that image is the outfit and the hand on dildo, the hand balancing is fucked

Anonymous

Anonymous Mon 27 Nov 2023 14:50:58 No.7677427 Report

Quoted By:

>>7677424
sure you do, joker

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 14:52:39 No.7677429 Report

Quoted By: >>7677440

>>7677416
This doesn't have a white background though?
>>7677419
Idk, I remember not being able to get walking/bouncing boobs without a lora

Anonymous

Anonymous Mon 27 Nov 2023 14:53:01 No.7677430 Report

Quoted By: >>7677431 >>7677437

>>7677426
>thats a basic pov shot which local can do fine
Including the hair grab, bank notes and spank marks?

>the only thing naiv3 did right in that image is the outfit and the hand on dildo
And the physics acting on hair and breasts and the face being upside down and the body not turning into flesh blobs, but sure.

Anonymous

Anonymous Mon 27 Nov 2023 14:53:38 No.7677431 Report

Quoted By: >>7677435 >>7677442

>>7677430
just inpaint bro

Anonymous

Anonymous Mon 27 Nov 2023 14:54:00 No.7677434 Report

Quoted By:

>>7677420
You could start by scraping and cleaning the data? That is what /lmg/ did with lewd language models. Anons downloaded, cleaned and shared datasets.
You don't need to be so entitled.
Also furry community and furry websites aren't so toxic so that helps also. e621 is really easy to scrape compared to Sankaku

Anonymous

Anonymous Mon 27 Nov 2023 14:54:05 No.7677435 Report

Quoted By:

>>7677431
Yes, that's why I asked "how long would it take".

Anonymous

Anonymous Mon 27 Nov 2023 14:54:25 No.7677436 Report

Quoted By: >>7677438

>>7677426
>the hand balancing is fucked
What does this even mean

Anonymous

Anonymous Mon 27 Nov 2023 14:54:45 No.7677437 Report

Quoted By:

>>7677430
for the former i'm sure you can jut inpaint them on in minutes

i wouldnt know about upside down shots because thats not what im interested in, idk, youre probably right

Anonymous

Anonymous Mon 27 Nov 2023 14:55:19 No.7677438 Report

Quoted By: >>7677445 >>7677449

>>7677436
look at the hand bro it has two thumbs and two palms

Anonymous

View Same Google ImgOps iqdb SauceNAO 3ca3e78c-cd4c-4f09-b3f9-3c06fa04 (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 14:55:48 No.7677439 Report

Quoted By:

>>7677376
given how small most anime tongues are, i don't think it's possible to balance an entire used hentai condom on them lol

Anonymous

Anonymous Mon 27 Nov 2023 14:56:16 No.7677440 Report

Quoted By: >>7677445

>>7677429
carried by artist and even then it's just kinda low-res

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{zero two (darling in the fran (...).png, 2MiB, 1536x1536

Anonymous Mon 27 Nov 2023 14:56:37 No.7677442 Report

Quoted By:

>>7677431
>just spend forever inpainting basically the whole image
Nah

Anonymous

Anonymous Mon 27 Nov 2023 14:57:04 No.7677443 Report

Quoted By: >>7677444

>>7677420
NAIv3 was trained on a fucking H100 cluster. The 4090 is a toy compared to that.

>>7677426
>the hand balancing is fucked
I only count 5 fingers, in 1.5 I would consider that a massive win.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1701091599314190.png, 56KiB, 252x147

Anonymous Mon 27 Nov 2023 14:58:38 No.7677444 Report

Quoted By: >>7677446 >>7677450

>>7677443
actually blind

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{kujikawa rise}}}, {{{persona (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 14:58:56 No.7677445 Report

Quoted By: >>7677448 >>7677483

>>7677438
Oh I only looked at the peach one
>>7677440
>carried by artist
??? If you could do that with a lora on local you'd be gushing about how amazing it is

Anonymous

Anonymous Mon 27 Nov 2023 14:59:46 No.7677446 Report

Quoted By:

>>7677444
That's an autumn leaf on top of the hand, not a finger :^)

Anonymous

Anonymous Mon 27 Nov 2023 15:00:29 No.7677448 Report

Quoted By: >>7677455

>>7677445
no lol

Anonymous

View Same Google ImgOps iqdb SauceNAO vladilena millize, {nyantcha}, m (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:01:23 No.7677449 Report

Quoted By: >>7677451 >>7677452

>>7677438
You're really acting like V1 was good with hands?
Hands are still iffy, but you quite literally have much more control with V3 than you ever had in V1 with hands.

Anonymous

Anonymous Mon 27 Nov 2023 15:02:09 No.7677450 Report

Quoted By: >>7677452

>>7677444
Are you? the thumb is rough, but it is clearly a single finger.

Anonymous

Anonymous Mon 27 Nov 2023 15:02:21 No.7677451 Report

Quoted By: >>7677460

>>7677449
Dude stop arguing with autists. You will NEVER convince them. Results don't matter, the ideology does.

Anonymous

Anonymous Mon 27 Nov 2023 15:02:49 No.7677452 Report

Quoted By:

>>7677449
no, i literally said that the hands were/are better in v3, illiterate

>>7677450
ecksdee

Anonymous

Anonymous Mon 27 Nov 2023 15:03:13 No.7677453 Report

Quoted By:

>>7677424
>A6000
my condolences

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{kujikawa rise}}}, {{{persona (...).png, 2MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:05:52 No.7677455 Report

Quoted By: >>7677456

>>7677448
I doubt
>>7677424
>autists act like this and local stays shit and never improves
>Novelai improves and releases a model that is a a huge step up
>Proceed to seethe at said model(even though you'd be slurping it up if it got leaked)
I don't get it

Anonymous

Anonymous Mon 27 Nov 2023 15:08:59 No.7677456 Report

Quoted By:

>>7677455
it's seething and coping
literally

Anonymous

View Same Google ImgOps iqdb SauceNAO 15.png, 1MiB, 1024x768

Anonymous Mon 27 Nov 2023 15:09:16 No.7677457 Report

Quoted By: >>7677461 >>7677462 >>7677463

>>7677426
shills btfo lmao

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{kujikawa rise}}}, {{{persona (...).png, 2MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:09:16 No.7677458 Report

Quoted By: >>7677459

>4
>Persona 4
Novelai is so fuarking SOUL

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, Yoshizawa Kasumi, persona (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:11:07 No.7677459 Report

Quoted By: >>7677465

>>7677458
I've barely gone into the SFW side, but I am really impressed with what it can do sometimes. I remember masks and stuff being an issue with V1, but that isn't the case anymore.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{artist_ fizrotart}}}, {lappla (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:12:21 No.7677460 Report

Quoted By:

>>7677451
this
do not feed the localoids

Anonymous

Anonymous Mon 27 Nov 2023 15:13:17 No.7677461 Report

Quoted By: >>7677477

>>7677457
False flag?

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{kujikawa rise}}}, {{{persona (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 15:13:33 No.7677462 Report

Quoted By: >>7677467

>>7677457
>four sets of hands
>fucked up hands
>boobs way too low on body
>dildo placement looks strange like it's not in where her pussy is supposed to be but stabbed into her
So glad I don't have to deal with that shit anymore. You can get similar issues on novelai SOMETIMES, it's the minority. But in local? That's the average result right there

Anonymous

Anonymous Mon 27 Nov 2023 15:13:34 No.7677463 Report

Quoted By:

>>7677457
>4 hands
Jogger in the first pic is getting fucked lol

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{kujikawa rise}}}, {{{persona (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:16:10 No.7677465 Report

Quoted By: >>7677467

Wife
>>7677459
Nice, it nails the persona style so well. Having so much fun with it

Anonymous

Anonymous Mon 27 Nov 2023 15:18:05 No.7677467 Report

Quoted By: >>7677478

>>7677462
>>7677465
Post some /h/

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:18:41 No.7677469 Report

Quoted By:

Localoid currently searching through his folder for a image that doesn't look like shit to post meanwhile novelSCHOLARS gen stuff like this in a minute tops

Anonymous

View Same Google ImgOps iqdb SauceNAO 1929e019-004c-432d-b5b8-aa335d33 (...).png, 2MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:19:12 No.7677472 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 15:19:43 No.7677473 Report

Quoted By: >>7677476

Please stop I don't want to pay

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, solo, {{akiyama rinko}}, (...).png, 3MiB, 1664x1216

Anonymous Mon 27 Nov 2023 15:20:35 No.7677475 Report

Quoted By:

>>7677387
Tried prompting the taimanin games he's worked on and now I'm getting much closer, thanks.

Anonymous

Anonymous Mon 27 Nov 2023 15:21:00 No.7677476 Report

Quoted By: >>7677482

>>7677473
Nigger that's why these faggot shills are here in the first place.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, brown hair, braid, red ey (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 15:21:05 No.7677477 Report

Quoted By: >>7677533

>>7677461
>no VAE
>base resolution
Of course it's bait. This consolewar-like shit flinging is mind numbing

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:21:40 No.7677478 Report

Quoted By: >>7677480

>>7677467

Anonymous

Anonymous Mon 27 Nov 2023 15:23:02 No.7677480 Report

Quoted By: >>7677484

>>7677478
Appreciated

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:25:41 No.7677482 Report

Quoted By:

>>7677476
LocalCONTRARIAN seething since he can't larp as some tech wizard artist or some shit just because he inpaints an entire image for hours on end, just to get MOGGED by novelGAWDS showing off what they prompted in just 30 seconds

Anonymous

Anonymous Mon 27 Nov 2023 15:26:13 No.7677483 Report

Quoted By: >>7677486 >>7677499

>>7677445
i just wish it could do p4's matsunaga ayana as well, can't have everything i guess

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 15:26:52 No.7677484 Report

Quoted By: >>7677494

>>7677480

Anonymous

Anonymous Mon 27 Nov 2023 15:27:16 No.7677486 Report

Quoted By:

>>7677483
matsunaga ayane

Anonymous

Anonymous Mon 27 Nov 2023 15:27:35 No.7677487 Report

Quoted By: >>7677493 >>7677510 >>7677534

I don't think NAI is good for me it satisfies the lazy part of me too well. Before I'd scour the web for a lora, grid out the strength to see what's best, sometimes even inpaint if I had a good image that could be great. Now I just click button. Had some plans for Lora's I was gonna make, some that I never finished, but now I feel no desire to do it.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, artist_spacezin, black ha (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:28:14 No.7677489 Report

Quoted By:

There is just too many artists to test.

Anonymous

Anonymous Mon 27 Nov 2023 15:32:01 No.7677493 Report

Quoted By:

>>7677487
>inb4 new popular character comes out and nai can't do it
It's going to be a vicious cycle

Anonymous

Anonymous Mon 27 Nov 2023 15:33:50 No.7677494 Report

Quoted By:

>>7677484
Very soft

Anonymous

Anonymous Mon 27 Nov 2023 15:35:22 No.7677496 Report

Quoted By:

I managed to get two patreon watermarks and one artist signature in three different corners of a random gen despite having them as UC tags
I hope this is something they fix in v4

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, {{{matsunaga ayane}}}, pe (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:36:20 No.7677499 Report

Quoted By: >>7677501

>>7677483
Is this kind of close? I think nailing her hair color is difficult.

Anonymous

Anonymous Mon 27 Nov 2023 15:38:27 No.7677500 Report

Quoted By: >>7677502 >>7677503 >>7677504

Anyone done any stuff with the characters of K-on! ?

Anonymous

Anonymous Mon 27 Nov 2023 15:38:54 No.7677501 Report

Quoted By:

>>7677499
kinda? i can see it but it just doesn't feel like her if you know what i mean. even among the p4 cast her design stood out as being different and "mobby" in a way

Anonymous

Anonymous Mon 27 Nov 2023 15:39:26 No.7677502 Report

Quoted By:

>>7677500
you missed some footjobs in the last thread

Anonymous

Anonymous Mon 27 Nov 2023 15:39:38 No.7677503 Report

Quoted By:

>>7677500
No, why would I !?

Anonymous

Anonymous Mon 27 Nov 2023 15:40:14 No.7677504 Report

Quoted By:

>>7677500
Yeah Mugi and Yui

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, solo, {artist_aoi n (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:40:43 No.7677505 Report

Quoted By: >>7677513 >>7677646

So close yet so far. Tried putting mizuki yukikaze in negatives but there's still some strange red coloring coming from something. Might still be better than my old Rinko lora's though

Anonymous

Anonymous Mon 27 Nov 2023 15:41:40 No.7677506 Report

Quoted By: >>7677507

>>7677342
hundreds

Anonymous

Anonymous Mon 27 Nov 2023 15:42:16 No.7677507 Report

Quoted By: >>7677508 >>7677512 >>7677521

>>7677506
Even with pov, all fours, bent over, from above
I get basic doggystyle shots

Anonymous

Anonymous Mon 27 Nov 2023 15:43:23 No.7677508 Report

Quoted By: >>7677515

>>7677507
I don't get what the difference is

Anonymous

View Same Google ImgOps iqdb SauceNAO 00202-1193997247.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 15:45:21 No.7677510 Report

Quoted By:

>>7677487
It just gets the job done. MAYBE I'll fix eyes once in a while but even then I can't be assed.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw,1girl,1boy, pov, pov, all f (...).png, 1014KiB, 832x1216

Anonymous Mon 27 Nov 2023 15:46:13 No.7677512 Report

Quoted By: >>7677515 >>7677523

>>7677507
I can't visualize what you want, how is it different from this?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, {{{akiyama rinko}}}, taim (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:46:23 No.7677513 Report

Quoted By:

>>7677505
Try reordering your tags and putting alternate costume into negs.

Anonymous

Anonymous Mon 27 Nov 2023 15:50:09 No.7677515 Report

Quoted By: >>7677516 >>7677518 >>7677525

>>7677512
>>7677508
No I mean I get a shot that isn't POV at all

Anonymous

Anonymous Mon 27 Nov 2023 15:52:00 No.7677516 Report

Quoted By: >>7677520

>>7677515
Post an example?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, vladilena millize, artist (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:54:58 No.7677518 Report

Quoted By: >>7677520

>>7677515
Do you mean something like this where you can see both characters?
The leg is fucked in this one, but both characters are visible.
I think I do understand what you want with the from above tag, but the from above tag is kind of fucked because the pictures in it aren't usually straight above the scene.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, sonia (Fire e (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:57:10 No.7677520 Report

Quoted By: >>7677523

>>7677516
>>7677518
Just took the first thing my current prompt gave me
I typically get something like this or >>7677518
Do I need to move the positioning tags earlier in the prompt?

Anonymous

Anonymous Mon 27 Nov 2023 15:57:21 No.7677521 Report

Quoted By: >>7677522 >>7677525

>>7677507
What about just from behind, all fours, back, back of head
Something like that

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, 1boy, yae miko, {{{from a (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 15:58:36 No.7677522 Report

Quoted By:

>>7677521

Anonymous

Anonymous Mon 27 Nov 2023 15:58:48 No.7677523 Report

Quoted By: >>7677524

>>7677520
Tag position matters a lot. Try importing the prompts from >>7677512 and >>7677372 and try following their order.

Anonymous

Anonymous Mon 27 Nov 2023 15:59:46 No.7677524 Report

Quoted By: >>7677546

>>7677523
nta but I feel like nsfw at the start degrades quality

Anonymous

Anonymous Mon 27 Nov 2023 15:59:49 No.7677525 Report

Quoted By:

>>7677515
I really doubt that's a problem on nai, even on local I just put in 1girl, 1boy, doggystyle, pov and get >>7677070 without any effort.

>>7677521
There is a doggystyle tag. imo much more reliable to do {doggystyle, pov} if you do have to force it, rather than putting in a bunch of tags that might fight each other for focus.

Anonymous

Anonymous Mon 27 Nov 2023 16:05:11 No.7677533 Report

Quoted By:

>>7677477
>This consolewar-like shit flinging is mind numbing
It's extremely fucking tiresome. Staber shitter, comfy wars, XL wars, now NAI wars, how fucking underage do you have to be to care about this shit so much

Anonymous

Anonymous Mon 27 Nov 2023 16:06:57 No.7677534 Report

Quoted By: >>7677537

>>7677487
>the convenient thing got more convenient and i don't like it
What kind of cope is this?

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, viewed from a unique angle (...).png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 16:07:57 No.7677536 Report

Quoted By:

>>7677278
Indeed

Anonymous

Anonymous Mon 27 Nov 2023 16:09:33 No.7677537 Report

Quoted By:

>>7677534
Same kind of cope we heard from /ic/.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1b3600da-d042-4008-ba0a-4ab340b9 (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 16:09:54 No.7677538 Report

Quoted By:

last little obscure gen before i head off

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{europa (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 16:10:02 No.7677539 Report

Quoted By: >>7677543 >>7677548 >>7677684

Europa

Anonymous

View Same Google ImgOps iqdb SauceNAO 00277-1471700585.png, 888KiB, 1152x768

Anonymous Mon 27 Nov 2023 16:12:23 No.7677542 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 16:12:27 No.7677543 Report

Quoted By: >>7677547 >>7677577

>>7677539
Need Heles or Metera fellatio gesture

Anonymous

View Same Google ImgOps iqdb SauceNAO 83-1girl,by Yoji Shinkawa,holdin (...).png, 2MiB, 1216x832

Anonymous Mon 27 Nov 2023 16:14:31 No.7677546 Report

Quoted By:

>>7677524
It definitely depends on the context, particularly artist tags. I've seen it both improve and degrade quality. I like using it at the start for simple sex prompts because it keeps you from needing to actually explicitly include a sex act in the prompt, so you get a variety of positions.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{europa (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 16:14:55 No.7677547 Report

Quoted By:

>>7677543
I'll give it a go with Metera

Anonymous

View Same Google ImgOps iqdb SauceNAO 00582-6155712803.png, 1MiB, 1280x1024

Anonymous Mon 27 Nov 2023 16:15:34 No.7677548 Report

Quoted By:

>>7677539

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{europa (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 16:17:07 No.7677550 Report

Quoted By: >>7677566

>Those hips
HOLY

Anonymous

Anonymous Mon 27 Nov 2023 16:18:38 No.7677552 Report

Quoted By: >>7677772

>>7677289
Post it

Anonymous

View Same Google ImgOps iqdb SauceNAO 00802-3880268897.png, 1MiB, 1280x1024

Anonymous Mon 27 Nov 2023 16:30:42 No.7677566 Report

Quoted By: >>7677569 >>7677570

>>7677550
so good!

Anonymous

View Same Google ImgOps iqdb SauceNAO 02268-6891614927.png, 2MiB, 1280x1024

Anonymous Mon 27 Nov 2023 16:35:22 No.7677569 Report

Quoted By:

>>7677566

Anonymous

Anonymous Mon 27 Nov 2023 16:35:41 No.7677570 Report

Quoted By: >>7677571 >>7677573

>>7677566
Djeeropa

Anonymous

Anonymous Mon 27 Nov 2023 16:36:37 No.7677571 Report

Quoted By:

>>7677570
squint harder

Anonymous

View Same Google ImgOps iqdb SauceNAO 01649-6317596721.png, 2MiB, 1280x1024

Anonymous Mon 27 Nov 2023 16:37:52 No.7677573 Report

Quoted By:

>>7677570

Anonymous

Anonymous Mon 27 Nov 2023 16:40:46 No.7677576 Report

Quoted By: >>7677582

>>7677042
The file is too large

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 16:41:09 No.7677577 Report

Quoted By: >>7677580 >>7677591 >>7677599

>>7677543

Anonymous

Anonymous Mon 27 Nov 2023 16:41:28 No.7677578 Report

Quoted By: >>7677586

>>7677370
uhm..

Anonymous

Anonymous Mon 27 Nov 2023 16:42:44 No.7677580 Report

Quoted By:

>>7677577
Nice
Best outfit too

Anonymous

Anonymous Mon 27 Nov 2023 16:43:03 No.7677581 Report

Quoted By:

>>7677383
that seems mostly like a you thing

Anonymous

Anonymous Mon 27 Nov 2023 16:43:27 No.7677582 Report

Quoted By: >>7677583

>>7677576
masterpiece,best quality,ultra-detailed:1.4), highres, 1girl, nsfw, <lora:add_detail:1>, <lora:character_pokemon_serena_v3:1>,serena $pokemon$, blue eyes, smile, hat, eyewear on headwear, high-waist skirt, thighhighs, nude
Negative prompt: (EasyNegative),(worst quality, low quality:1.3), simple background, public hair, blurry, bad-hands-5, badhandv4, verybadimagenegative_v1.3, extra fingers:1.5, watermark, signature, extra arms, extra hands, extra legs, extra limbs, extra body parts, extra breasts, mangled fingers, (bad body structure:1.2), badly drawn hands,missing arms, missing breasts, missing extremities, missing hands, missing limbs, missing lips, mutated hands, (mutation:1.2), poorly drawn hands, logo, futanari, geometry, painting, sketch,osaic censoring, censored,
Steps: 20, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 417968067, Size: 2624x3968, Model hash: a49140c6c5, Model: mistoonAnime_v10, Denoising strength: 0.29, Clip skip: 2, Ultimate SD upscale upscaler: 4x-UltraSharp, Ultimate SD upscale tile_width: 512, Ultimate SD upscale tile_height: 512, Ultimate SD upscale mask_blur: 8, Ultimate SD upscale padding: 32, CFG Rescale phi: 0, Lora hashes: "add_detail: 7c6bad76eb54, character_pokemon_serena_v3: 665bdbbaa236", Version: v1.4.1

Anonymous

Anonymous Mon 27 Nov 2023 16:44:43 No.7677583 Report

Quoted By: >>7677596 >>7677601

>>7677582
I see that from the png info but, like I said how did it generate reverse cowgirl from any of that?

Anonymous

Anonymous Mon 27 Nov 2023 16:46:23 No.7677584 Report

Quoted By:

>>7677383
I don't even care that local is higher effort, I am fine with doing inpaints and editingng shit myself, and in fact do that with SFW.
It's just that it's almost impossible to make the style appealing to me, especially if you aren't doing a simple pinup or a landscape. It's always glossy and shiny unless you use some very overfit lora that will make prompting sex even more fucking impossible. The bodies all look fucking same. I am sick and tired of aom belly and ribs. I can't look at this shit without feeling disgust already.

Anonymous

Anonymous Mon 27 Nov 2023 16:46:23 No.7677585 Report

Quoted By: >>7677587

>>7677420
They are laughing at us because they opened a discord server and started to organize properly so people could contribute with images and tagging for the dataset instead of all of them being a selfish piece of shit who just talks garbage, thinks that the world and everything is against them, being bitter agains eachother, and waits for things to happen magically. I really am disgusted of furfags and their gay zoophile garbage but I am very impressed about what they did together, using a single 3090.

Anonymous

Anonymous Mon 27 Nov 2023 16:47:06 No.7677586 Report

Quoted By: >>7677593

>>7677578
What? Sorry, is the style not to your liking? Would you like some AOM2 slop instead?

Anonymous

Anonymous Mon 27 Nov 2023 16:49:10 No.7677587 Report

Quoted By:

>>7677585
Furfaggots used jewgle TPU gibs they literally give TPUs to anyone

Anonymous

View Same Google ImgOps iqdb SauceNAO 02902-6595269343.png, 2MiB, 1280x1024

Anonymous Mon 27 Nov 2023 16:52:29 No.7677591 Report

Quoted By:

>>7677577
ok you win
im surprised nai3 can get the facial expressions so good

Anonymous

Anonymous Mon 27 Nov 2023 16:54:11 No.7677593 Report

Quoted By: >>7677598

>>7677586
>style
you do you gay guy

Anonymous

Anonymous Mon 27 Nov 2023 16:54:37 No.7677596 Report

Quoted By: >>7677613

>>7677583
nta, that metadata is worthless because it was taken from img2img (you can tell because it contains "Denoising strength"), likely used for an upscale. It's not the original txt2img metadata.

Anonymous

Anonymous Mon 27 Nov 2023 16:56:20 No.7677598 Report

Quoted By: >>7677600 >>7677608 >>7677672

>>7677593
i bet you generate niggers fucking your waifu lmfao

Anonymous

Anonymous Mon 27 Nov 2023 16:58:06 No.7677599 Report

Quoted By: >>7677607

>>7677577
Are you using the half-closed eyes tag? It's good

Anonymous

Anonymous Mon 27 Nov 2023 16:59:30 No.7677600 Report

Quoted By: >>7677610 >>7677615

>>7677598
didn't expect to side with shoebill but please curbstomp localoid copers harder

Anonymous

Anonymous Mon 27 Nov 2023 16:59:38 No.7677601 Report

Quoted By:

>>7677583
nta either, but if you ask for nsfw and do not specify a position you will still get one

Anonymous

View Same Google ImgOps iqdb SauceNAO 00821-5537537325.png, 2MiB, 1280x1024

Anonymous Mon 27 Nov 2023 17:00:53 No.7677603 Report

Quoted By:

question is can nai3 do epic big bois?

Anonymous

Anonymous Mon 27 Nov 2023 17:01:34 No.7677604 Report

Quoted By: >>7677613 >>7677624

>>7676932
Why do his balls have nipples

Anonymous

Anonymous Mon 27 Nov 2023 17:01:38 No.7677605 Report

Quoted By: >>7677621

My only complaint with V3 is that for some reason the pregnant bellies look like balloons more often than not though the navels aren't fucked up anymore so there's that at least

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 17:02:11 No.7677607 Report

Quoted By:

>>7677599
No I was just using
Fellatio gesture
Oral invitation
Open mouth, tongue out is also good but I didn't use it for that

Anonymous

Anonymous Mon 27 Nov 2023 17:02:32 No.7677608 Report

Quoted By:

>>7677598
nah but nice of you for telling us about your fetish I guess

Anonymous

Anonymous Mon 27 Nov 2023 17:03:28 No.7677610 Report

Quoted By: >>7677631

>>7677600
Oh come on I'm not THAT recognizable.

Anonymous

Anonymous Mon 27 Nov 2023 17:04:45 No.7677613 Report

Quoted By:

>>7677596
thanks, I was wondering if I was missing something. Is there a way to import my tiled diffusion>region prompt control prompts from png info? Trying to fine tune and figure this out more keep getting images blending or coming out weird with extra characters.
https://files.catbox.moe/4r8hda.jpg
https://files.catbox.moe/0tfkke.png
>>7677604
maybe it feels good to have them pinched?

Anonymous

Anonymous Mon 27 Nov 2023 17:05:29 No.7677615 Report

Quoted By:

>>7677600
>localoid copers
>when all the nigger gens are NAIv3
lolmao

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, morgan_le_fay (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 17:06:26 No.7677617 Report

Quoted By: >>7677660 >>7677681 >>7677694

Takeuchi is kinda scuffed

Anonymous

View Same Google ImgOps iqdb SauceNAO illyasviel von einzbern, fate, { (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 17:06:28 No.7677618 Report

Quoted By: >>7677660

artist:santa (sunflower)

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 17:06:59 No.7677620 Report

Quoted By: >>7677641

Love Metera

Anonymous

View Same Google ImgOps iqdb SauceNAO 02797-7055384542.png, 2MiB, 1280x1024

Anonymous Mon 27 Nov 2023 17:07:12 No.7677621 Report

Quoted By: >>7677632

>>7677605
there is a lora that solved this one, has been out for sometime on civitai

Anonymous

Anonymous Mon 27 Nov 2023 17:07:34 No.7677622 Report

Quoted By: >>7677630 >>7677634

This is nice for instant gratification / easy comprehension of some artists, but imma just burn all my leftover Anlas from a year ago and then finally get off my ass and learn how to train a LORA instead, it's fun to use once but I can't justify paying for it with enough hardware to run local.

Anonymous

View Same Google ImgOps iqdb SauceNAO 108-holding gun 1boy povart by w (...).png, 2MiB, 1824x1248

Anonymous Mon 27 Nov 2023 17:08:05 No.7677624 Report

Quoted By:

>>7677604
Inserting nipples onto anything vaguely rounded is just what 1.5 does.

Upscaling cleavage can be a nightmare, even with (nipples:1.3) in the negatives, it still slaps them on half the time or more.

Anonymous

Anonymous Mon 27 Nov 2023 17:09:12 No.7677629 Report

Quoted By: >>7677636 >>7677643 >>7677647 >>7677655 >>7677689

Could someone try artist Obui?

Anonymous

Anonymous Mon 27 Nov 2023 17:09:32 No.7677630 Report

Quoted By: >>7677657

>>7677622
>spending all the time to train loras for worse results

Anonymous

Anonymous Mon 27 Nov 2023 17:09:44 No.7677631 Report

Quoted By: >>7677649

>>7677610
>fizrot
>brings up niggers
you are lol

Anonymous

Anonymous Mon 27 Nov 2023 17:10:00 No.7677632 Report

Quoted By:

>>7677621
I know about the 20 pregnant loras, I had the one called "pregnantonly" to use it under clothes. V3 solved the navel nipple but the pov pregnant bellies look a bit too round. Could be the style I'm using

Anonymous

Anonymous Mon 27 Nov 2023 17:10:12 No.7677633 Report

Quoted By:

>>7677391
>>7677383
I've been thinking about getting a second PC just to gen images on, would make local alot more viable with style loras

Anonymous

Anonymous Mon 27 Nov 2023 17:10:14 No.7677634 Report

Quoted By: >>7677657

>>7677622
post if you find an easy guide

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 17:10:45 No.7677636 Report

Quoted By: >>7677641

>>7677629
Sure, one sec

Anonymous

Anonymous Mon 27 Nov 2023 17:12:04 No.7677641 Report

Quoted By:

>>7677636
>>7677620
metera sexo

Anonymous

View Same Google ImgOps iqdb SauceNAO 4857389457398.png, 1MiB, 1024x1536

Anonymous Mon 27 Nov 2023 17:12:09 No.7677642 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00950-5754781815.png, 2MiB, 1280x1024

Anonymous Mon 27 Nov 2023 17:12:33 No.7677643 Report

Quoted By:

>>7677629
>artist obui
sadge

Anonymous

View Same Google ImgOps iqdb SauceNAO 389475893475893.png, 1MiB, 1024x1536

Anonymous Mon 27 Nov 2023 17:12:48 No.7677644 Report

Quoted By: >>7677677

Anonymous

Anonymous Mon 27 Nov 2023 17:13:08 No.7677646 Report

Quoted By: >>7677651 >>7677764

>>7677505
is this the taimanin schizo that kept posting basic NAI looking taimanin images for a year or so?
It's amazing it still kinda looks like basic NAI even on v3.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 2MiB, 1280x1856

Anonymous Mon 27 Nov 2023 17:13:17 No.7677647 Report

Quoted By: >>7677658

>>7677629

Anonymous

Anonymous Mon 27 Nov 2023 17:13:24 No.7677649 Report

Quoted By:

>>7677631
... touché.

Anonymous

View Same Google ImgOps iqdb SauceNAO 389457839475389.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 17:13:27 No.7677650 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 17:13:52 No.7677651 Report

Quoted By: >>7677660

>>7677646
What the fuck I had completely forgotten that
VAE NOW

Anonymous

Anonymous Mon 27 Nov 2023 17:14:42 No.7677655 Report

Quoted By:

>>7677629
Fauxbui generally works but the AI can't replicate his shading so it ends up looking worse

Anonymous

Anonymous Mon 27 Nov 2023 17:15:58 No.7677657 Report

Quoted By: >>7677666 >>7677671 >>7677673

>>7677630
>Spending all your money to gen images using SECRET SAUCE and learning nothing
I can use hyperbole too, nigger
>>7677634
Once I get off my ass and gain momentum I have a high tolerance for figuring this shit out, but I don't expect to find an "easy" way.

Tbh if I didn't enjoy tinkering NAI would be the way to go, not because I think shills/doubters are correct that NAI is unequivocally better, but it is undeniably easier, and for some people I can understand the reasons for paying for convenience, esp. if they don't have the hardware already.

Anonymous

Anonymous Mon 27 Nov 2023 17:16:08 No.7677658 Report

Quoted By:

>>7677647
Thanks. That's really nice for a quick prompt.

Anonymous

Anonymous Mon 27 Nov 2023 17:16:16 No.7677659 Report

Quoted By: >>7677745

>>7676998
Got more Reimus totally naked and with no head-wear?

Anonymous

Anonymous Mon 27 Nov 2023 17:16:19 No.7677660 Report

Quoted By:

>>7677618
how about "tsuji santa"?

>>7677617
Yeah I dunno how they managed to make Takeuchi look this shit

>>7677651
>VAE NOW
im gonna miss this...

Anonymous

Anonymous Mon 27 Nov 2023 17:18:26 No.7677666 Report

Quoted By:

>>7677657
>once I get off my ass and gain momentum
the fact that you prefaced it like that means it will never happen

Anonymous

Anonymous Mon 27 Nov 2023 17:18:37 No.7677667 Report

Quoted By: >>7677692

>>7677101
That's what the leaker said as well as one Nai staff so that is what there is to go on.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{europa (granblue fantasy)}}}, (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 17:19:10 No.7677671 Report

Quoted By:

>>7677657
>has to dance around saying novelai is just plain out higher quality
Tinker all you want, you won't get stuff like this on local

Anonymous

View Same Google ImgOps iqdb SauceNAO e8afa99aa8992f44013d2a6372d0c354.png, 3MiB, 1447x1437

Anonymous Mon 27 Nov 2023 17:19:26 No.7677672 Report

Quoted By: >>7677678 >>7677686

>>7677598
>nai3 promoters can't beat proompters with their tool so they resort to insecure racial slurs
>based twitter moment

make sure to screencap, tag nai and hit the like and subscribe button

Anonymous

Anonymous Mon 27 Nov 2023 17:19:42 No.7677673 Report

Quoted By:

>>7677657
Sounds like it should just be a black box, you stick some tagged images and a base model in one end and it pops out a lora on the other end.

Anonymous

Anonymous Mon 27 Nov 2023 17:21:00 No.7677677 Report

Quoted By:

>>7677644
what nai3 model is this

Anonymous

Anonymous Mon 27 Nov 2023 17:21:12 No.7677678 Report

Quoted By: >>7677685

>>7677672
I haven't seen any good local stuff posted this thread, it ALL has that shitty same slop style

Anonymous

Anonymous Mon 27 Nov 2023 17:21:16 No.7677679 Report

Quoted By:

>>7677101
>fags are still pushing the github 0day theory
>casually just breached github, a microsoft owned enterprise platform for an anime model and not God knows what corporate secrets
>"lol no serious company actually uses github"
jesus christ

Anonymous

Anonymous Mon 27 Nov 2023 17:21:45 No.7677680 Report

Quoted By: >>7677689

why do you retards still reply to the same bait after like a week?
brain damage.

Anyways, if danbooru had fucking CGs dumped the quality would skyrocket, shame it's so incredibly cucked and also gives in to artist blacklist demands

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, 1boy, artoria pendragon ( (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 17:22:30 No.7677681 Report

Quoted By: >>7677702

>>7677617
Looks kinda alright though?

Anonymous

Anonymous Mon 27 Nov 2023 17:22:46 No.7677682 Report

Quoted By: >>7677687

So how come no one claimed responsibility for NAI leak yet? Did he not get fired either - does no one know who did it?

Anonymous

Anonymous Mon 27 Nov 2023 17:22:57 No.7677683 Report

Quoted By: >>7677701

Any good loras for 'nasty' women? I mean the style of Neisan, Iku, etc.
I have two neisan and iku loras and both suck ass

Anonymous

View Same Google ImgOps iqdb SauceNAO 15314-709037249-masterpiece, 1gi (...).png, 829KiB, 640x1024

Anonymous Mon 27 Nov 2023 17:23:26 No.7677684 Report

Quoted By: >>7677729

>>7677539
I made an europa lora ages ago

Anonymous

Anonymous Mon 27 Nov 2023 17:23:37 No.7677685 Report

Quoted By:

>>7677678
you must be new here, have a (you)

Anonymous

Anonymous Mon 27 Nov 2023 17:23:45 No.7677686 Report

Quoted By: >>7677783

>>7677672
>haha le epic niggers are based GET OWN NAIOID
Irony-poisoned faggots like you should never poison the gene pool.

Anonymous

Anonymous Mon 27 Nov 2023 17:24:49 No.7677687 Report

Quoted By:

>>7677682
>does no one know who did it?
It was a github zero-day. The person who did it will never be found.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, sonia (Fire e (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 17:24:59 No.7677689 Report

Quoted By: >>7677833

>>7677629
Not curating these gens at all, but in terms of at least giving you the McDonalds variant of obui it seems close. The trademark big ass nipples/areolas and hyper masculine male build are there.

>>7677680
Yeah, that's why so much of sian's art isn't available and I hate it

Anonymous

Anonymous Mon 27 Nov 2023 17:25:57 No.7677692 Report

Quoted By:

>>7677667
NAI hosted their code on github, you can see the `.git` folders in the leak, you load those up in a git GUI to see that they're legit. The folder contains leftover github OAuth tokens, so those are legit too.
Unless the bulletproof fucking Github had a 0day, which would have devastated every single large software/tech firm in the world, doubt. More realistically, one of the nai devs had a virus on their machine that leaked their git repo access tokens or similar, and the files were just siphoned right off of their machine or from their online repository using their credentials. Could a lack of token validation have been the 0day? Possibly, but this attack was absolutely not some kind of blind attack on Github, that's insane, that sort of vulnerability would have rocked the entire internet, not even the most clout-chasing hacker would burn such a valuable vulnerability on cartoon porn, every single security professional and nation state would be salivating at the thought of turning in that bounty or weaponising it.
Even more realistically, some dude working for nai had full repo access and just slurped up the entire orgs files. What's interesting is that the files were all .tar -> .zstd archive compression. This is totally non-standard, most people use zip/7z/rar or something, this was totally out of nowhere. The use of such a compression format is not "scene friendly", this is something someone in a startup would do, using a brand new compression tech, that's why I firmly believe it was an inside job.
Still, it's just a fun theory, I don't give a shit either way unless we get a round two, and that ain't happening.

Anonymous

Anonymous Mon 27 Nov 2023 17:26:14 No.7677694 Report

Quoted By: >>7677700

>>7677617
have you tried with year xxxx

Anonymous

Anonymous Mon 27 Nov 2023 17:27:33 No.7677696 Report

Quoted By: >>7677711

Ways of tagging and Lora training are still very subjective and uncertain? How You guys haven't figured out a good way of thinking to improve tagging and training?

Anonymous

Anonymous Mon 27 Nov 2023 17:28:28 No.7677700 Report

Quoted By:

>>7677694
I have not

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 17:28:59 No.7677701 Report

Quoted By: >>7677703 >>7677709

>>7677683
Don't have any issues with neisans style

Anonymous

Anonymous Mon 27 Nov 2023 17:29:22 No.7677702 Report

Quoted By:

>>7677681
What the fuck
I wouldn't believe its ai if not the thread and larger saber tits

Anonymous

Anonymous Mon 27 Nov 2023 17:29:48 No.7677703 Report

Quoted By: >>7677715 >>7677719

>>7677701
Yeah but I'm not using NAI3. Any good loras for it?

Anonymous

Anonymous Mon 27 Nov 2023 17:30:42 No.7677706 Report

Quoted By:

You know how you can ctrl/cmd+click an image in your history sidebar and copy the prompt? Is there a way to do it on mobile? I have to upload the image every time otherwise.

Anonymous

Anonymous Mon 27 Nov 2023 17:31:05 No.7677709 Report

Quoted By: >>7677714

>>7677701
It actually does. A lot of the time Neisan in prompt will fuck up the anatomy HARD. Sometimes it doesn't and augments the nsfw quite a bit. Probably depends on the character and other artists in prompt

Anonymous

Anonymous Mon 27 Nov 2023 17:31:46 No.7677711 Report

Quoted By: >>7677721

>>7677696
no reason to think about that stuff anymore just use nai

Anonymous

Anonymous Mon 27 Nov 2023 17:32:21 No.7677714 Report

Quoted By: >>7677720

>>7677709
neisan himself will sometimes fuck the anatomy hard

Anonymous

Anonymous Mon 27 Nov 2023 17:32:23 No.7677715 Report

Quoted By: >>7677717

>>7677703
There's a civit lora that doesn't turn everyone into flesh blobs at high weights but it looks a bit off.
desu I preferred combining two gayshit loras at low weight

Anonymous

View Same Google ImgOps iqdb SauceNAO 00221-1955876662.png, 746KiB, 712x1072

Anonymous Mon 27 Nov 2023 17:33:01 No.7677716 Report

Quoted By:

>>7674053
I use ad detailer with hands lora in the hand detection prompt

Anonymous

Anonymous Mon 27 Nov 2023 17:33:05 No.7677717 Report

Quoted By: >>7677720

>>7677715
mix merge them and produce a new lora

Anonymous

Anonymous Mon 27 Nov 2023 17:33:44 No.7677718 Report

Quoted By:

>>7677140
>ComfyUI Manager Menu
has an option to download Auto1111 type nodes as well as downloading missing nodes from workflows
>Efficiency nodes
If you just wanna prompt it puts all the nodes you really need into just 2
You can also just try using the webui extension that adds comfy to auto since it just puts it into the pipeline which is mostly what I do if I wanna use comfy since even with the auto compatibility nodes I still don't care for comfyui handling the base gen.

Anonymous

Anonymous Mon 27 Nov 2023 17:33:47 No.7677719 Report

Quoted By: >>7677728

>>7677703
Didn't try it myself but https://civitai.com/models/139678 seems good.

Anonymous

Anonymous Mon 27 Nov 2023 17:34:26 No.7677720 Report

Quoted By:

>>7677714
Yeah that's true lol.
>>7677717
I mixed them at different weights depending on what I want so meh

Anonymous

Anonymous Mon 27 Nov 2023 17:36:19 No.7677721 Report

Quoted By: >>7677722 >>7677723 >>7677732

>>7677711
You really took it as a challenge to say something as stupid as possible, didn't you? Better ways of tagging and improved training could lead us to train a better NAIv3 to use on local

Anonymous

View Same Google ImgOps iqdb SauceNAO 1697916506860391.gif, 3MiB, 498x469

Anonymous Mon 27 Nov 2023 17:37:26 No.7677722 Report

Quoted By: >>7677730

>>7677721
>Better ways of tagging and improved training could lead us to train a better NAIv3 to use on local

Anonymous

Anonymous Mon 27 Nov 2023 17:38:41 No.7677723 Report

Quoted By: >>7677730

>>7677721
lol
perhaps even lmao

Anonymous

Anonymous Mon 27 Nov 2023 17:39:57 No.7677728 Report

Quoted By:

>>7677719
Looks nice. Really wish there was some lora capable of recreating Neisans/Iku messy hairy pussy and asshole style

Anonymous

View Same Google ImgOps iqdb SauceNAO Lake&Lyria.jpg, 3MiB, 2894x4093

Anonymous Mon 27 Nov 2023 17:40:34 No.7677729 Report

Quoted By: >>7677742

>>7677684
>>7677403
No Lyria?

Anonymous

Anonymous Mon 27 Nov 2023 17:40:43 No.7677730 Report

Quoted By: >>7677734 >>7677736 >>7677738 >>7677743 >>7677747

>>7677722
>>7677723
Come the fuck on. Dunk at "JUST INPAINT BRO" delusional retards all you want but laughing at any effort to do a better local model is just sure fucking grim.

Anonymous

Anonymous Mon 27 Nov 2023 17:41:23 No.7677732 Report

Quoted By: >>7677739 >>7677749

>>7677721
NTA but man, local did slack off an entire year after the NAIV1 leak, investing everything into loras only and not getting a better base model.

And people now are still ignoring SDXL finetuning for anime and now WE know how good it can be by what NAIv3 can do.

Local is fucked, for anime at least.

Anonymous

Anonymous Mon 27 Nov 2023 17:41:28 No.7677734 Report

Quoted By: >>7677748

>>7677730
What the fuck are you expecting of literal shills?

Anonymous

Anonymous Mon 27 Nov 2023 17:42:26 No.7677735 Report

Quoted By: >>7677746 >>7678059

>>7676927
sorry to ask but isn't there a chance that pixiv will ban you for copying some other pixiv artist's style? or do they not care (like at all)?

Anonymous

Anonymous Mon 27 Nov 2023 17:42:32 No.7677736 Report

Quoted By: >>7677748

>>7677730
once the hardware requirements aren't pegged at $10k+ for one 80GB GPU (at msrp, more like 2x with scalping), you will see more xl finetunes. Until then, local is 1.5.
It's not a lack of effort, the field is completely priced out due to insane stockpiling.

Anonymous

Anonymous Mon 27 Nov 2023 17:42:45 No.7677738 Report

Quoted By: >>7677748

>>7677730
Local didn't do shit all this time dude

Anonymous

Anonymous Mon 27 Nov 2023 17:43:07 No.7677739 Report

Quoted By: >>7677749

>>7677732
>local did slack off an entire year after the NAIV1 leak
You clearly were not here during the initial leak, what the fuck are you talking about? Local improved by tenfold after NAI leaked because things like loras, hypernets, and much more got reverse engineered.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{artist_fishine}}}, 3girls, li (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 17:43:14 No.7677740 Report

Quoted By: >>7677744

If you know you know.
(couldn't bother getting them in the right order)

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 17:43:37 No.7677742 Report

Quoted By: >>7677776

>>7677729
Doing Metera and voyeur stuff right now

Anonymous

Anonymous Mon 27 Nov 2023 17:43:43 No.7677743 Report

Quoted By: >>7677748 >>7677751

>>7677730
You will have better luck on /sdg/. There are a few finetuners still kicking and the pixart training is going on too. Seeing people offer training gpus here first is a joke. This is clearly a shill thread now

Anonymous

Anonymous Mon 27 Nov 2023 17:43:52 No.7677744 Report

Quoted By: >>7677768

>>7677740
there is futa cock, isn't there

Anonymous

Anonymous Mon 27 Nov 2023 17:44:09 No.7677745 Report

Quoted By: >>7677750

>>7677659
>>7677239
sorry done for today

Anonymous

Anonymous Mon 27 Nov 2023 17:44:26 No.7677746 Report

Quoted By: >>7677762

>>7677735
You can't make money off AI art on fanbox, that's about it. As long as it's properly tagged as AI you should be fine.

That said the rules keep changing, and getting copies this accurate was rare before naiv3.

Anonymous

Anonymous Mon 27 Nov 2023 17:44:50 No.7677747 Report

Quoted By: >>7677752

>>7677730
ill stop laughing once you fags actually do something aside from whine how no one is doing anything

Anonymous

Anonymous Mon 27 Nov 2023 17:45:12 No.7677748 Report

Quoted By: >>7677758

>>7677734
Well according to the schizos I am a "shill" as well.
>>7677736
Even improving 1.5 models could do good. Properly tagged dataset improves model capability by a fuckload. Yes it won't be as good as NAI3 but at least it won't be this goyslop we have now.
>>7677738
hdg took the lora approach I guess. This general produced quite a lot of loras. We lack just a single autist that would organize this like with furries...
>>7677743
>This is clearly a shill thread now
God I'm so tired of you fucktards. Fuck off.

Anonymous

Anonymous Mon 27 Nov 2023 17:45:49 No.7677749 Report

Quoted By:

>>7677732
>>7677739
you slacked off and all the aom2 shitmixes are the proof. once b64 came out you stopped trying at all save for some autists who insisted on autistic block merges and got practically nothing out of it

Anonymous

Anonymous Mon 27 Nov 2023 17:45:52 No.7677750 Report

Quoted By: >>7677753 >>7677775

>>7677745
it's fine anon, we both know sideways/upside down on 1.5 sucks. source: me, I rotate the canvas to inpaint.

Anonymous

Anonymous Mon 27 Nov 2023 17:46:02 No.7677751 Report

Quoted By: >>7677765

>>7677743
/sdg/ is a shitshow of avatarfags and shilling for ComfyUi.

Anonymous

Anonymous Mon 27 Nov 2023 17:46:32 No.7677752 Report

Quoted By:

>>7677747
This, guarantee in a year nothing will have changed

Anonymous

Anonymous Mon 27 Nov 2023 17:47:27 No.7677753 Report

Quoted By:

>>7677750
>me, I rotate the canvas to inpaint
yep that was fun though lol

Anonymous

View Same Google ImgOps iqdb SauceNAO 1686461150661415.jpg, 121KiB, 1068x544

Anonymous Mon 27 Nov 2023 17:47:45 No.7677755 Report

Quoted By:

>itt:

Anonymous

Anonymous Mon 27 Nov 2023 17:49:10 No.7677758 Report

Quoted By: >>7677761 >>7677767

>>7677748
>>This is clearly a shill thread now
>God I'm so tired of you fucktards. Fuck off.
Do you need 50 more replies from nai users insulting any effort you are putting in or are you ready to actually do something? Posting about training a finetune here is a waste of time when there are people actually making an effort instead of fellating a furry model that shits up your mix more often than not and sucks ass at native highres like it was advertised it wouldn't be.

Anonymous

Anonymous Mon 27 Nov 2023 17:50:43 No.7677761 Report

Quoted By:

>>7677758
trust me bro the furry model is good bro they did it all by themselves bro they have so much new tech bro

take the knot up your ass bro

Anonymous

Anonymous Mon 27 Nov 2023 17:50:52 No.7677762 Report

Quoted By:

>>7677746
yeah I want to use some artists but kind of afraid because it looks like their style really fucking closely

Anonymous

View Same Google ImgOps iqdb SauceNAO 00006-2620257735.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 17:51:05 No.7677764 Report

Quoted By:

>>7677646
Nope I'm the one who angrily replied to him to stop smearing Rinko. I was the guy who'd ask people to help with my rinko lora every now and then. This is one of my old posts, also using a schuffed hirasawa seiji lora

Anonymous

Anonymous Mon 27 Nov 2023 17:51:14 No.7677765 Report

Quoted By: >>7677769

>>7677751
Only place to discuss lora training atm, everything gets buried in shitflinging here

Anonymous

Anonymous Mon 27 Nov 2023 17:51:21 No.7677767 Report

Quoted By: >>7677770 >>7677799

>>7677758
It's just trolls shitting back for calling anyone who dares to acknowledge NAI's good sides a "shill". I can barely hold myself from shitposting back. It's fucking tiresome.

Anonymous

Anonymous Mon 27 Nov 2023 17:51:49 No.7677768 Report

Quoted By:

>>7677744
Correct.

Anonymous

Anonymous Mon 27 Nov 2023 17:52:29 No.7677769 Report

Quoted By: >>7677774

>>7677765
Hello? Have you looked at absolute state of /sdg/ now? Their shitflinging is even more disgusting with all this eceleb circlejerking.

Anonymous

Anonymous Mon 27 Nov 2023 17:52:45 No.7677770 Report

Quoted By:

>>7677767
I really don't give a shit about shills. Figure out who is doing the pixart shit and support him if you want something done. Point still stands, /h/ is stagnant af

Anonymous

Anonymous Mon 27 Nov 2023 17:52:50 No.7677771 Report

Quoted By: >>7677773

nai will start limiting the use of artist tags so they could only be used by accounts created before 1st january 2024

Anonymous

Anonymous Mon 27 Nov 2023 17:52:51 No.7677772 Report

Quoted By: >>7677779

>>7677552
i need to finish upgrading it

Anonymous

Anonymous Mon 27 Nov 2023 17:53:36 No.7677773 Report

Quoted By:

>>7677771
falseflaggot

Anonymous

Anonymous Mon 27 Nov 2023 17:54:17 No.7677774 Report

Quoted By:

>>7677769
This thread is shitflinging on accelerate what's your point?

Anonymous

Anonymous Mon 27 Nov 2023 17:55:38 No.7677775 Report

Quoted By: >>7677777 >>7677786

>>7677750
i never tried any prompts for it, if i really wanted it i could train it and you know it

Anonymous

Anonymous Mon 27 Nov 2023 17:56:21 No.7677776 Report

Quoted By:

>>7677742
Need some Metera presenting spread legs on the bed

Anonymous

Anonymous Mon 27 Nov 2023 17:56:34 No.7677777 Report

Quoted By: >>7677781

>>7677775
>training for upside-down on 1.5
Oh you sweet summer child

Anonymous

View Same Google ImgOps iqdb SauceNAO 22186-48796593.png, 3MiB, 1280x1600

Anonymous Mon 27 Nov 2023 17:56:43 No.7677778 Report

Quoted By: >>7677780 >>7677803

Anonymous

Anonymous Mon 27 Nov 2023 17:57:40 No.7677779 Report

Quoted By: >>7677782

>>7677772
I can wait. Post it when you're ready. I hope You don't forget to do it.

Anonymous

Anonymous Mon 27 Nov 2023 17:57:57 No.7677780 Report

Quoted By: >>7677785

>>7677778
Where's her cun

Anonymous

Anonymous Mon 27 Nov 2023 17:58:10 No.7677781 Report

Quoted By:

>>7677777
based quints and true words as well, nai is our saviour from dalle3

Anonymous

Anonymous Mon 27 Nov 2023 17:58:37 No.7677782 Report

Quoted By: >>7677788 >>7677826

>>7677779
i didn't spend thanksgiving weekend diving through 250 pages of the "intercourse with machinery" tag on pixiv to just leave it on the floor

Anonymous

Anonymous Mon 27 Nov 2023 17:59:14 No.7677783 Report

Quoted By: >>7677789

>>7677686
you sound upset, you should tell that to your dad instead

Anonymous

Anonymous Mon 27 Nov 2023 17:59:38 No.7677785 Report

Quoted By:

>>7677780
omitted for authentic sketchy watercolor experience

Anonymous

Anonymous Mon 27 Nov 2023 17:59:58 No.7677786 Report

Quoted By: >>7677790

>>7677775
I don't think you understand what you're talking about. Perhaps you'd make some amount of progress with native trainig or a finetune, but loras sure as hell fuck this concept up. Actually, there's a ton of things that loras just won't learn or learn poorly. For example specialy shaped pupils, upside down (or in-lap pov), feet/heels up in the air while lying down. I could go on anon, I've trained a ton of loras, I'm quite familiar with them.

Anonymous

Anonymous Mon 27 Nov 2023 18:00:38 No.7677788 Report

Quoted By:

>>7677782
Great to hear, fellow locals we need to ally ourselves

Anonymous

Anonymous Mon 27 Nov 2023 18:00:50 No.7677789 Report

Quoted By: >>7677808

>>7677783
Too bad, my dad wasn't absent or in prison unlike the average nigglett's

Anonymous

Anonymous Mon 27 Nov 2023 18:00:57 No.7677790 Report

Quoted By: >>7677793 >>7677794

>>7677786
i literally trained a heart shaped pupils lora though

Anonymous

Anonymous Mon 27 Nov 2023 18:03:28 No.7677793 Report

Quoted By: >>7678034

>>7677790
Is it a diff lora, or does it generalise the concept? Does it work when genning a full image? Are there any other attributes being learned, for example with a full character? If it's zoomed in on the eyes and only that, yeah of course it'd work. Makima eyes for example have their own inpainting-specific lora. There's even a cross-shaped inpainting lora, I'm talking about raw gens though. Unless you're reasonably close to the face, the concept doesn't work. At that point it's more an inpainting lora than something you can mix in.

Anonymous

Anonymous Mon 27 Nov 2023 18:03:43 No.7677794 Report

Quoted By:

>>7677790
yeah and it changes the rest of the image along with it

Anonymous

Anonymous Mon 27 Nov 2023 18:04:29 No.7677795 Report

Quoted By: >>7677809 >>7677940

Dude if I had a 4090 I really wouldn't care to learn for months how to train a finetune or a model from scratch I love all this crap and I like the idea of rebel against companies

Anonymous

Anonymous Mon 27 Nov 2023 18:04:41 No.7677797 Report

Quoted By:

Furries literally have local models that prompts furshit as good as nai3 does anime, plus all the added control of local. It’s clearly possible to just make a new thing

Anonymous

Anonymous Mon 27 Nov 2023 18:04:59 No.7677799 Report

Quoted By: >>7677805 >>7677814

>>7677767
initially when people started to talk about it and say how it was an improvement at least upon conceptual understanding, it did sound interesting, but if it had any good points it sure got shat out by the shilling

Anonymous

View Same Google ImgOps iqdb SauceNAO naiv3testingu.png, 1MiB, 1088x1600

Anonymous Mon 27 Nov 2023 18:05:47 No.7677803 Report

Quoted By:

>>7677778
cozy style

Anonymous

Anonymous Mon 27 Nov 2023 18:06:01 No.7677805 Report

Quoted By: >>7677811 >>7677821

>>7677799
Yeah because the shilling clearly affects how good it is, the more you shill the worse the anatomy gets

In real time.

Anonymous

Anonymous Mon 27 Nov 2023 18:06:58 No.7677808 Report

Quoted By: >>7677813

>>7677789
which says a lot about you

thank you for promoting NAI3 sir

Anonymous

Anonymous Mon 27 Nov 2023 18:07:00 No.7677809 Report

Quoted By: >>7677818

>>7677795
based if poster. i believe you, trust me.

Anonymous

Anonymous Mon 27 Nov 2023 18:07:08 No.7677811 Report

Quoted By:

>>7677805
It will revert back to AOM2 soon if we keep shilling it

Anonymous

View Same Google ImgOps iqdb SauceNAO {artist_saitou masatsugu}, 1girl (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 18:08:05 No.7677812 Report

Quoted By: >>7677815 >>7677819

One thing I've noticed that when you prompt in artists who really only do worksafe stuff that you're going to run into an issue with nipples.

Anonymous

Anonymous Mon 27 Nov 2023 18:08:31 No.7677813 Report

Quoted By: >>7677829

>>7677808
Sorry that your dad abandoned you or whatever, hope you get over it localoid

Anonymous

Anonymous Mon 27 Nov 2023 18:08:33 No.7677814 Report

Quoted By: >>7677842

>>7677799
>but if it had any good points it sure got shat out by the shilling
What? How the fuck does it work exactly?

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 2girls, 1boy, scathach_ska (...).png, 3MiB, 1856x1280

Anonymous Mon 27 Nov 2023 18:09:09 No.7677815 Report

Quoted By: >>7677838

>>7677812
Wasn't an issue for me with Koyama's style

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 18:11:09 No.7677817 Report

Quoted By: >>7677822 >>7677851

Anonymous

Anonymous Mon 27 Nov 2023 18:11:23 No.7677818 Report

Quoted By:

>>7677809
Yes I post here often (not images tho), and thanks, I actually liked the process of training a Lora, it's a lof of fun for me

Anonymous

Anonymous Mon 27 Nov 2023 18:11:52 No.7677819 Report

Quoted By: >>7677832

>>7677812
Try moving nsfw to the front.

Anonymous

Anonymous Mon 27 Nov 2023 18:12:25 No.7677821 Report

Quoted By:

>>7677805
yeah I don't think most people enjoy being gaslit into using a shitty service that they even have to pay for

Anonymous

Anonymous Mon 27 Nov 2023 18:12:34 No.7677822 Report

Quoted By: >>7677823

>>7677817
Inpaint those panties out so it's just the thong nigga

Anonymous

Anonymous Mon 27 Nov 2023 18:13:11 No.7677823 Report

Quoted By:

>>7677822
Nah it's one of those fancy hybrids

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 18:14:36 No.7677825 Report

Quoted By: >>7677834 >>7677851

Combining multiple views and anime screencap seems to do a good job

Anonymous

Anonymous Mon 27 Nov 2023 18:15:17 No.7677826 Report

Quoted By: >>7678035

>>7677782
Also lad have You trained a character recently? What important settings and amount of images for dataset do You recommend for it?

Anonymous

Anonymous Mon 27 Nov 2023 18:15:44 No.7677829 Report

Quoted By:

>>7677813
you're projecting, but that's nothing new from shillfags

Anonymous

Anonymous Mon 27 Nov 2023 18:16:34 No.7677832 Report

Quoted By: >>7677836

>>7677819
Yeah. That seemed to fix it. Kind of hate how NAI's own guide suggests that any kind of quality tag goes to the end while they straight up say the most important things should be at the front.

Anonymous

Anonymous Mon 27 Nov 2023 18:16:34 No.7677833 Report

Quoted By:

>>7677689
Yeah, even though the face is off all his main features are there.

Anonymous

Anonymous Mon 27 Nov 2023 18:16:49 No.7677834 Report

Quoted By:

>>7677825
Surprised it's keeping the outfit as consistent as it is to be honest

Anonymous

Anonymous Mon 27 Nov 2023 18:18:26 No.7677836 Report

Quoted By:

>>7677832
It's not a quality tag and it's at the front in the negative, so it makes sense that it should go at the front for the regular prompt too.

Anonymous

Anonymous Mon 27 Nov 2023 18:18:38 No.7677838 Report

Quoted By: >>7677841 >>7677843

>>7677815
can they do standing sex

Anonymous

View Same Google ImgOps iqdb SauceNAO Untitled.png, 2MiB, 1216x832

Anonymous Mon 27 Nov 2023 18:20:40 No.7677841 Report

Quoted By: >>7677843

>>7677838
I had better gens than this one for standing that I forgot to save before ending my tab/session yesterday, this is the only one I kept it looks like
Whoops.

Anonymous

View Same Google ImgOps iqdb SauceNAO artist_abmayo, nsfw, hatsune mik (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 18:20:48 No.7677842 Report

Quoted By: >>7677847 >>7677850 >>7677852

>>7677814
Mental illness, don't feed the localoids

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 2girls, 1boy, scathach_ska (...).png, 3MiB, 1856x1280

Anonymous Mon 27 Nov 2023 18:23:03 No.7677843 Report

Quoted By: >>7677845

>>7677841
>>7677838
Also setting up proper 3Ps with penetration is a pain in the ass and I don't recommend it.
EIther the bodies get fucked up or the hands get fucked up or it wants the hand to be a dick or it decides both girls need to be getting fucked even if you have 2boys in undesired etc.

Anonymous

Anonymous Mon 27 Nov 2023 18:24:38 No.7677845 Report

Quoted By:

>>7677843
Don't even try standing missionary

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, Koyanskaya (F (...).png, 2MiB, 1024x1024

Anonymous Mon 27 Nov 2023 18:24:46 No.7677846 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 18:25:01 No.7677847 Report

Quoted By: >>7677849

>>7677842
uh schizo that seems like some pedo shit

Anonymous

Anonymous Mon 27 Nov 2023 18:26:07 No.7677849 Report

Quoted By: >>7677864

>>7677847
>hatsune miku is pedo shit
This might be the funniest localoid cope yet.

Anonymous

Anonymous Mon 27 Nov 2023 18:26:28 No.7677850 Report

Quoted By: >>7677852 >>7677856

>>7677842
It's kinda sad that it has come to this though, wonder if it's going to be possible just to post shit without those nutjobs shitting up the thread.

Anonymous

Anonymous Mon 27 Nov 2023 18:27:33 No.7677851 Report

Quoted By: >>7677853 >>7677858 >>7677869

>>7677817
>>7677825
can you do granblue fantasy sex?

Anonymous

Anonymous Mon 27 Nov 2023 18:28:54 No.7677852 Report

Quoted By:

>>7677842
>>7677850
the nai3 gang showed up and started shitting on localoids saying thats how it started

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, metera, Granb (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 18:29:35 No.7677853 Report

Quoted By: >>7677862

>>7677851
NTA but I did some, I was just hot swapping artists at the time so my prompt/artist accuracy isn't as good as anon's

Anonymous

Anonymous Mon 27 Nov 2023 18:30:13 No.7677854 Report

Quoted By:

localoid cope through the roof

Anonymous

Anonymous Mon 27 Nov 2023 18:30:57 No.7677856 Report

Quoted By: >>7677861

>>7677850
same as always, anon arrives with some good points but is so fucking bad at communicating that everyone hates the guy

nai3 is amazing, but it ain't local. I'm glad anon is happy with his gens, wish he wasn't such a cunt about it though, just enjoy genning and post cool stuff.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, Heles (Granbl (...).png, 2MiB, 1280x1856

Anonymous Mon 27 Nov 2023 18:32:13 No.7677858 Report

Quoted By: >>7677862

>>7677851
I also gave Mushi024 a try

Anonymous

Anonymous Mon 27 Nov 2023 18:34:08 No.7677861 Report

Quoted By: >>7677864

>>7677856
The problem is that localoids have acquired a sense of artoid-like pride on top of the standard insufferable FOSSfag pride (even though V1 is a proprietary model, leak or not) and teasing them is too much fun.

Anonymous

Anonymous Mon 27 Nov 2023 18:34:46 No.7677862 Report

Quoted By: >>7677913 >>7677918

>>7677858
this looks good, if only could be colored
>>7677853
its coherent but meh

Anonymous

Anonymous Mon 27 Nov 2023 18:36:03 No.7677864 Report

Quoted By: >>7677873

>>7677849
>h-haha guys it's not like that
sure sounds like some naifag copium

>>7677861
>sweaty keyboard clicking intensifies

Anonymous

View Same Google ImgOps iqdb SauceNAO 00479-1872868815.png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 18:37:36 No.7677865 Report

Quoted By: >>7677868

This whole NAIV3 vs Local thing is just a lesser flavor of doing things yourself vs using tools. We've come full circle.

Anonymous

Anonymous Mon 27 Nov 2023 18:40:29 No.7677868 Report

Quoted By:

>>7677865
Except those tools are 1000% better

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 18:41:17 No.7677869 Report

Quoted By:

>>7677851
In a bit ill gen some

Anonymous

View Same Google ImgOps iqdb SauceNAO DEEP HURTING, nsfw,1girl, {{1boy (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 18:41:39 No.7677871 Report

Quoted By: >>7677872 >>7677877 >>7677879 >>7677906 >>7677911 >>7677913 >>7677952 >>7678004

What saddens me is that most of the local posters have mostly given up posting anything, all that's left is low effort trolls on that side. It would be kind of fun if there was actually an attempt at competition, instead of this endless embarrassing shitflinging.

Anonymous

Anonymous Mon 27 Nov 2023 18:42:26 No.7677872 Report

Quoted By: >>7677874

>>7677871
This general was literally on its deathbed for weeks before nai.

Anonymous

View Same Google ImgOps iqdb SauceNAO anchovy (girls und panzer), pov, (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 18:42:39 No.7677873 Report

Quoted By: >>7677878 >>7677900

>>7677864
Cope and seethe.

Anonymous

Anonymous Mon 27 Nov 2023 18:43:14 No.7677874 Report

Quoted By: >>7677876

>>7677872
People got bored looking at AOM slop

Anonymous

View Same Google ImgOps iqdb SauceNAO nnnnnnnn.jpg, 416KiB, 832x1216

Anonymous Mon 27 Nov 2023 18:43:44 No.7677875 Report

Quoted By: >>7677968

Anonymous

Anonymous Mon 27 Nov 2023 18:44:34 No.7677876 Report

Quoted By: >>7678004

>>7677874
That sure fucking applies to me. I got to genning sfw where it was somewhat possible to make it not look like aomsloppa

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1536x1536

Anonymous Mon 27 Nov 2023 18:44:56 No.7677877 Report

Quoted By: >>7677899

>>7677871
What tag did you use for the panties slid over like that bro?

Anonymous

Anonymous Mon 27 Nov 2023 18:45:06 No.7677878 Report

Quoted By:

>>7677873
indeed you do

Anonymous

Anonymous Mon 27 Nov 2023 18:45:56 No.7677879 Report

Quoted By: >>7677919

>>7677871
I keep posting, I've stopped getting replies though. I'm still having fun.

Anonymous

Anonymous Mon 27 Nov 2023 18:46:21 No.7677881 Report

Quoted By: >>7677883 >>7677884 >>7677888 >>7677896

Local is Linux and NAI is Mac(donalds)

Anonymous

Anonymous Mon 27 Nov 2023 18:47:39 No.7677883 Report

Quoted By:

>>7677881
>local is troonix
Judging by the localoids' attitude that's spot on

Anonymous

Anonymous Mon 27 Nov 2023 18:48:09 No.7677884 Report

Quoted By:

>>7677881
Has anyone trained a dobson Lora yet?

Anonymous

Anonymous Mon 27 Nov 2023 18:48:58 No.7677886 Report

Quoted By: >>7677890

still not buying your online service

Anonymous

Anonymous Mon 27 Nov 2023 18:49:16 No.7677888 Report

Quoted By:

>>7677881
lul

Anonymous

Anonymous Mon 27 Nov 2023 18:49:51 No.7677890 Report

Quoted By: >>7677901

>>7677886
Post your local gens then

Anonymous

Anonymous Mon 27 Nov 2023 18:50:48 No.7677896 Report

Quoted By:

>>7677881
is nai american?

Anonymous

View Same Google ImgOps iqdb SauceNAO e71516f3-a398-4a10-be2c-734ec13e (...).png, 820KiB, 832x1216

Anonymous Mon 27 Nov 2023 18:52:23 No.7677899 Report

Quoted By:

>>7677877
Clearly it was 'sexual sex in the sex postion for sex involving a penis and a vagina' (the typo in position is important).

Or in other words pure Gacha

Anonymous

View Same Google ImgOps iqdb SauceNAO anchovy (girls und panzer), pov, (...).png, 2MiB, 832x1216

Anonymous Mon 27 Nov 2023 18:52:38 No.7677900 Report

Quoted By: >>7677905 >>7677912 >>7677917

>>7677873
>ai-generated tag really gives you AOM slop
My sides are in orbit

Anonymous

Anonymous Mon 27 Nov 2023 18:53:19 No.7677901 Report

Quoted By:

>>7677890
But I am. I'm just not avatarfagging.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yaia (granblue fantasy)}}}, { (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 18:55:43 No.7677905 Report

Quoted By:

>>7677900
LMAAAAOOO

Anonymous

Anonymous Mon 27 Nov 2023 18:55:48 No.7677906 Report

Quoted By: >>7677915

>>7677871
Wouldn't call it a competition when the only judges are the participants themselves. AOM "slop" is new, it's what I'm here for. Anime has been around for what, 50 years? Don't see the point in trying to mimic it now.

That said I'd rather interact with the few localfags left than fling shit at NAI's side and have shit flung back.

Anonymous

Anonymous Mon 27 Nov 2023 18:57:18 No.7677911 Report

Quoted By:

>>7677871
i dont care either way but one thing is for sure, im glad i didnt release any tech
anons who gen waifus are insufferable

Anonymous

Anonymous Mon 27 Nov 2023 18:57:51 No.7677912 Report

Quoted By:

>>7677900
anyone interested in joining the nai3 gang should follow this fine gentleman's pics

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, Metera, Granb (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 18:57:58 No.7677913 Report

Quoted By: >>7677951

>>7677862
>its coherent but meh
I'm sure you could make it significantly better if you actually built the prompt for it
>>7677862
>this looks good, if only could be colored
You can, I set the prompt to monochrome and greyscale.
In my experience it looked significantly worse if you try to direct the AI to color it because it struggles to replicate Mushi's shading as it is and when you make it color it it becomes even worse at it. Maybe there's prompt magic to make it work better but idk

>>7677871
The AI really fucked her bangs on this one, looks like she did the fusion dance with Pholia

Anonymous

Anonymous Mon 27 Nov 2023 18:59:07 No.7677915 Report

Quoted By: >>7677923

>>7677906
bro its an anime board go to aco or something if youre in desperate need for pseudorealistic dog vomit

Anonymous

Anonymous Mon 27 Nov 2023 18:59:16 No.7677917 Report

Quoted By: >>7677928

>>7677900
>no ribs
FAKE

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 18:59:30 No.7677918 Report

Quoted By: >>7677929

>>7677862
Masturbation

Anonymous

Anonymous Mon 27 Nov 2023 18:59:31 No.7677919 Report

Quoted By:

>>7677879
I still enjoy looking at the local styles but i don't reply to them.

Anonymous

Anonymous Mon 27 Nov 2023 19:02:18 No.7677923 Report

Quoted By: >>7677932

>>7677915
Have you ever been to /aco/'s AI thread?

Anonymous

View Same Google ImgOps iqdb SauceNAO ai-generated, anchovy (girls und (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 19:05:36 No.7677928 Report

Quoted By: >>7677934

>>7677917
Better? :)

Anonymous

Anonymous Mon 27 Nov 2023 19:05:41 No.7677929 Report

Quoted By: >>7677938

>>7677918
>double moles
ruined

Anonymous

Anonymous Mon 27 Nov 2023 19:07:00 No.7677932 Report

Quoted By:

>>7677923
nope not interested

Anonymous

Anonymous Mon 27 Nov 2023 19:07:23 No.7677934 Report

Quoted By: >>7677936

>>7677928
Oh my god the SOVL

Anonymous

Anonymous Mon 27 Nov 2023 19:08:39 No.7677936 Report

Quoted By: >>7677950 >>7678004

>>7677934
I know you're baiting but just give it some time and seething contrarian localoids will unironically call AOM slop sovl

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 19:08:56 No.7677938 Report

Quoted By:

>>7677929
Fixed lol

Anonymous

View Same Google ImgOps iqdb SauceNAO {ai-generated}, 1girl, hakurei r (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 19:09:19 No.7677939 Report

Quoted By: >>7677949

Anonymous

Anonymous Mon 27 Nov 2023 19:09:50 No.7677940 Report

Quoted By:

>>7677795
training loras is fun ig, but finetuning an entire model, specially SDXL, is either not possible on a 4090 because not enough VRAM or it will take ages.

I think NAIv1 was made on an A100 cluster, v3 was finetuned on an H100 cluster.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 19:13:51 No.7677947 Report

Quoted By: >>7677964

Anonymous

Anonymous Mon 27 Nov 2023 19:16:26 No.7677949 Report

Quoted By:

>>7677939
needed a double take to make sue this ain't AOM. kek that stomach is too perfect

Anonymous

Anonymous Mon 27 Nov 2023 19:17:01 No.7677950 Report

Quoted By:

>>7677936
something something the nostalgia for a past form of media is always centered on its flaws

Anonymous

View Same Google ImgOps iqdb SauceNAO DEEP HURTING, nsfw,1girl, {{1boy (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 19:17:08 No.7677951 Report

Quoted By: >>7677964

>>7677913
Is this better? I have no knowledge of the character or game whatsoever, I just lazily copied the other anon's prompt because I liked the borders and style.

Anonymous

Anonymous Mon 27 Nov 2023 19:17:26 No.7677952 Report

Quoted By: >>7677960

>>7677871
People won't get anything better with pure txt2img on local tho, and if they post something good people will say "that took ages, aom shit, etc"

I enjoyed local a lot, for an entire year and training loras and such, but I can't be blind and say naiv3 is bad when it gets the job done with just prompting. It's just better for now.

It's still missing features I would like to have that local does, but without them it just generated better images anyways.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{metera (granblue fantasy)}}}, (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 19:19:26 No.7677957 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 19:20:01 No.7677960 Report

Quoted By:

>>7677952
You can post 1girls in pretty sceneries like some anons still do. With enough styling and correct loras you can make something that is nice to look at. Local model hentai looks absolutely inferior though, can't do much with that.

Anonymous

Anonymous Mon 27 Nov 2023 19:20:41 No.7677962 Report

Quoted By: >>7677979

reminder that controlnet WILL come to NAI V3 in case you miss having a reason to fuck with blender

Anonymous

Anonymous Mon 27 Nov 2023 19:21:17 No.7677964 Report

Quoted By:

>>7677951
Yeah, that's slightly better. Her bangs typically form a slight V shape at the end like >>7677947 rather than being flat across

Anonymous

View Same Google ImgOps iqdb SauceNAO b5380543-b1fb-4a87-b0cd-3a31d39d (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 19:22:48 No.7677967 Report

Quoted By: >>7677970

when are we going to get 3d to 2d pipeline? thats the next big step from txt2image because it would give us the ability to make actual comics. just place the models and render it as a drawing

Anonymous

View Same Google ImgOps iqdb SauceNAO 9c3d8b6b15b4ae30c56896076c349cc2 (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 19:23:24 No.7677968 Report

Quoted By: >>7677971 >>7677982 >>7679776

>>7677875
Inpaint anon with NAIv3? Looks really nice. Poor abmayo, lol.

Anonymous

Anonymous Mon 27 Nov 2023 19:23:47 No.7677970 Report

Quoted By: >>7677973 >>7677981

>>7677967
controlnet does that already

Anonymous

Anonymous Mon 27 Nov 2023 19:23:58 No.7677971 Report

Quoted By:

>>7677968
he inpainted previously posted nai3 gen

Anonymous

Anonymous Mon 27 Nov 2023 19:24:11 No.7677972 Report

Quoted By: >>7677975 >>7677976 >>7677977

Looking at AOM slop for more than two seconds brought up an interesting question: what the fuck did they fine-tune it on?
The AOM default style looks like gook/chink shit (leaning towards gook) and not Japanese art

Anonymous

View Same Google ImgOps iqdb SauceNAO a623b384-dbed-4ed0-87b4-ba712e17 (...).png, 1015KiB, 832x1216

Anonymous Mon 27 Nov 2023 19:24:50 No.7677973 Report

Quoted By: >>7677978 >>7677990

>>7677970
not really what i mean. controlnet abstracts away everything but the pose, im talking keeping the whole scene

Anonymous

Anonymous Mon 27 Nov 2023 19:25:20 No.7677975 Report

Quoted By: >>7677988

>>7677972
3D instagram models, no?

Anonymous

Anonymous Mon 27 Nov 2023 19:25:30 No.7677976 Report

Quoted By: >>7677988

>>7677972
AOM is a block merge.
https://huggingface.co/WarriorMama777/OrangeMixs#abyssorangemix2-aom2

Anonymous

Anonymous Mon 27 Nov 2023 19:25:52 No.7677977 Report

Quoted By: >>7677988

>>7677972
https://huggingface.co/WarriorMama777/OrangeMixs

BMW with a realistic model

Anonymous

Anonymous Mon 27 Nov 2023 19:25:57 No.7677978 Report

Quoted By:

>>7677973
to be more specific i mean that you'd be able to feed in metadata between 3d and 2d, like telling it "ok these polygons belong to character X"

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, {artist_spacezin}, 1girl, (...).jpg, 369KiB, 1280x1856

Anonymous Mon 27 Nov 2023 19:27:39 No.7677979 Report

Quoted By: >>7677991

>>7677962
I already fuck with Blender enough but I do look forward to that

Anonymous

Anonymous Mon 27 Nov 2023 19:30:06 No.7677981 Report

Quoted By: >>7677986 >>7677996

>>7677970
Controlnet is pure 2D, you can't do things like characters overlapping without a major headache, an equivalent trained on the concept of 3d models could be very interesting. though it would probably need to be trained into the model itself, and require a completely modified architecture.

Anonymous

Anonymous Mon 27 Nov 2023 19:31:11 No.7677982 Report

Quoted By: >>7677984 >>7677997 >>7678003

>>7677968
I don't have nai, local for life unfortunately. I can certainly pay for it, but once I get one hit I'm dealthly afraid that I'd get addicted.
man nai composition is fucking sick isn't it
I can't wait for xl finetunes, please spoil me with more mikus

Anonymous

Anonymous Mon 27 Nov 2023 19:32:07 No.7677984 Report

Quoted By: >>7677993

>>7677982
>dealthly afraid that I'd get addicted
that's a very valid concern. it feels like I am cooking my brain with this.

Anonymous

Anonymous Mon 27 Nov 2023 19:33:22 No.7677986 Report

Quoted By: >>7677989 >>7677996

>>7677981
>he doesn't know about segmentation and compositing
>he doesn't know you can automate the process in comfyui

Anonymous

Anonymous Mon 27 Nov 2023 19:34:33 No.7677988 Report

Quoted By: >>7677992 >>7677994

>>7677975
>>7677976
>>7677977
>it was basilmix all along
Why? I never bothered to look into it but WHY?

Anonymous

Anonymous Mon 27 Nov 2023 19:35:25 No.7677989 Report

Quoted By: >>7678000

>>7677986
All of that is way more trouble than it is worth, rarely worth the effort, and not remotely comparable to what I was talking about.

Anonymous

Anonymous Mon 27 Nov 2023 19:35:31 No.7677990 Report

Quoted By: >>7677996

>>7677973
One controlnet for lineart, created from your 3D image by an edge detect filter, like laplace. Another controlnet for poses, using the 3d skeleton. Then use the actual 3D image as a base for img2img, guided by those chained controlnets.

Anonymous

Anonymous Mon 27 Nov 2023 19:35:40 No.7677991 Report

Quoted By:

>>7677979
drunk woman my beloved

Anonymous

Anonymous Mon 27 Nov 2023 19:36:22 No.7677992 Report

Quoted By:

>>7677988
I think it's because it a 3dpd model that was advertised to have tagging that has danbooru tagging. when I just started proompting I mixed Basilmix into my meme mix exactly because of this lol and I didn't know about aom

Anonymous

Anonymous Mon 27 Nov 2023 19:36:51 No.7677993 Report

Quoted By:

>>7677984
I was endlessly training prior to V3 so I feel like just prompting has been a break for my brain instead of just squinting at different epochs and trying to figure out what one was best, or tweaking training settings and training over and over again.

Anonymous

Anonymous Mon 27 Nov 2023 19:37:30 No.7677994 Report

Quoted By: >>7678015

>>7677988
3dpd

also prob (or hope) most anons don't about x-taber models, but some people tried to do things with that model and nai leaked model

Anonymous

View Same Google ImgOps iqdb SauceNAO z21.png, 818KiB, 832x1216

Anonymous Mon 27 Nov 2023 19:38:10 No.7677996 Report

Quoted By:

>>7677981
yeah thats what i mean. i've tried with converting renders into 2d and it gets fucked once the poses and interactions become non-generic
https://www.reddit.com/r/StableDiffusion/comments/181il6p/3dcontrolled_video_generation_with_live_previews/
it seems like some people are working on it, but the example given is fairly trivial with base SD
>>7677986
>>7677990
im sure its possible but at that point you may as well draw it yourself

Anonymous

Anonymous Mon 27 Nov 2023 19:38:17 No.7677997 Report

Quoted By:

>>7677982
>spoil me with more mikus
any style preferences? not promising any results though

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, {{{{{{very aesthetic}}}}}} (...).png, 268KiB, 360x600

Anonymous Mon 27 Nov 2023 19:38:28 No.7677999 Report

Quoted By:

artist more like fartist lmao gottem
naishills btfo

Anonymous

Anonymous Mon 27 Nov 2023 19:39:41 No.7678000 Report

Quoted By:

>>7677989
Then it's that 3d model dataset maker ani has on the backburner

Anonymous

View Same Google ImgOps iqdb SauceNAO 4fb412fa7f1d92c25cff185a0ca1498f (...).png, 910KiB, 832x1216

Anonymous Mon 27 Nov 2023 19:41:04 No.7678003 Report

Quoted By: >>7678008 >>7678012

>>7677982
I know how you feel. I said fuck it and sent you a key anyway though since I appreciate your contributions here a lot. You should know where to look. If you decide not to redeem it for whatever reason I'll still post some Miku here regardless.

Anonymous

Anonymous Mon 27 Nov 2023 19:41:47 No.7678004 Report

Quoted By:

>>7677871
>most of the local posters have mostly given up posting anything
Or just taking their time/doing other shit. The only things I've worked on in the past day are non-H.
https://files.catbox.moe/qn3xb0.png
https://files.catbox.moe/zvs9ap.png
https://files.catbox.moe/oqnsfk.png
(as usual don't bother checking metadata you'll just get photoshop)
>>7677876
You could make porn look distinctly different too, though?
Literal skill/effort issue.
>>7677936
AOM shit never looked good. It was the mark of laziness on the part of the user.
And I wouldn't pretend default NAI looks all that much better. It overly flattens the fuck out of a lot of things and has a lot of inconsistent linework. And the fucking eyes give are so bad it physically hurts.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 2girls, eonsang , navel, b (...).png, 969KiB, 832x1216

Anonymous Mon 27 Nov 2023 19:41:53 No.7678005 Report

Quoted By:

>>7676928
>how do you get that level of quality?
wgmi

Anonymous

View Same Google ImgOps iqdb SauceNAO 9a052617-0143-409f-8170-9e804171 (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 19:43:36 No.7678006 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 19:44:19 No.7678008 Report

Quoted By: >>7678122

>>7678003
Could you please give a hint how to obtain those keys? I can't pay for NAI directly.

Anonymous

Anonymous Mon 27 Nov 2023 19:46:04 No.7678010 Report

Quoted By: >>7678014

Fuck off beggars
No begging allowed
Go flip some burgers

Anonymous

Anonymous Mon 27 Nov 2023 19:46:34 No.7678012 Report

Quoted By: >>7678024

>>7678003
rocking the boat I see

Anonymous

Anonymous Mon 27 Nov 2023 19:47:12 No.7678014 Report

Quoted By:

>>7678010
I don't mind paying money though, not begging.

Anonymous

Anonymous Mon 27 Nov 2023 19:47:20 No.7678015 Report

Quoted By: >>7678018

>>7677994
Unfortunately I know about those, I was here when the C one was released by the glowniggers

Anonymous

Anonymous Mon 27 Nov 2023 19:49:44 No.7678018 Report

Quoted By: >>7678019

>>7678015
Why would they release the best realistic model (Cstaber) to the public?

Anonymous

Anonymous Mon 27 Nov 2023 19:51:18 No.7678019 Report

Quoted By:

>>7678018
Falseflagging, baiting, etc
You know how crafty and dishonest the kikes are.

Anonymous

Anonymous Mon 27 Nov 2023 19:53:32 No.7678021 Report

Quoted By: >>7678025 >>7678028 >>7678044

Funny how there's no artist//ic/ outrage against NAIv3 yet when it can faithfully replicate artist styles even without LoRAs

Anonymous

View Same Google ImgOps iqdb SauceNAO b0d5411fb3610d704015bcf2908465c0 (...).png, 926KiB, 832x1216

Anonymous Mon 27 Nov 2023 19:53:45 No.7678024 Report

Quoted By: >>7678031

>>7678012
hope you find your brush ;)

Anonymous

Anonymous Mon 27 Nov 2023 19:54:30 No.7678025 Report

Quoted By:

>>7678021
Probably because it is pay to use.

Anonymous

Anonymous Mon 27 Nov 2023 19:58:23 No.7678028 Report

Quoted By:

>>7678021
pay2use i guess, also maybe they will notice it eventually when people do porn with their style instead of paying them (damn that would be fun)

Anonymous

Anonymous Mon 27 Nov 2023 19:58:38 No.7678029 Report

Quoted By: >>7678042

The AI seems to be able to get certain characters with far fewer entries than I initially thought, sometimes even as few as like 90-120 is enough

Anonymous

Anonymous Mon 27 Nov 2023 20:01:17 No.7678031 Report

Quoted By:

>>7678024
based

Anonymous

Anonymous Mon 27 Nov 2023 20:02:25 No.7678034 Report

Quoted By:

>>7677793
it's a specific inpaint lora yeah but training refuses to pick up on tiny details like that otherwise, please let me know if there's a breakthrough on that

Anonymous

Anonymous Mon 27 Nov 2023 20:03:06 No.7678035 Report

Quoted By:

>>7677826
no and as many as possible

Anonymous

Anonymous Mon 27 Nov 2023 20:03:11 No.7678036 Report

Quoted By: >>7678054 >>7678125

If only there was a way to get the super close shot NAIv3 quality eyes on stuff other than PoV paizuri/blowjob shots

Anonymous

Anonymous Mon 27 Nov 2023 20:04:54 No.7678039 Report

Quoted By:

I'm new, what's the usual go-to settings for Image2Image? Like strength, noise.

Anonymous

Anonymous Mon 27 Nov 2023 20:07:07 No.7678041 Report

Quoted By:

>>7677129
>>7677288
>>7677307
>>7677293
>upload a bunch of missing art under the WRONG artist name (if they're blacklisted)
>NovelAI takes snapshot
>Danbooru is more complete

Would this work or no?

Anonymous

Anonymous Mon 27 Nov 2023 20:07:35 No.7678042 Report

Quoted By: >>7678048

>>7678029
Is it only trained on danbooru? No other booru?

Anonymous

Anonymous Mon 27 Nov 2023 20:08:00 No.7678044 Report

Quoted By: >>7678049 >>7678052 >>7678054

>>7678021
>when it can faithfully replicate artist styles even without LoRAs
It really can't, though. Shading and color grading is usually pretty off and linework is inconsistent at best. It can produce a facsimile but still feels a lot like an off-brand. And the ones it does actually get fairly close to are both limited to being nips(limiting sphere of impact) and pretty specific ones at that, so there isn't much to care about unless you're one of them.
That said, I never understood the fascination of wanting to copy an artist style 1:1. Linework, shading, general palette use? I could understand wanting to pick and choose bits and pieces from an artist but just directly replicating their shit is so lazy and uninspired.

Anonymous

Anonymous Mon 27 Nov 2023 20:11:47 No.7678048 Report

Quoted By:

>>7678042
As far as I know, yes

Anonymous

Anonymous Mon 27 Nov 2023 20:13:20 No.7678049 Report

Quoted By: >>7678051 >>7678071

>>7678044
"I wish this artist drew the character I liked"
or
"I like this artist but they only draw NTR"
or
"I like this artist but they don't draw porn"
Basically it comes from those places.

Anonymous

Anonymous Mon 27 Nov 2023 20:13:51 No.7678051 Report

Quoted By:

>>7678049
this desu senpai

Anonymous

Anonymous Mon 27 Nov 2023 20:14:05 No.7678052 Report

Quoted By: >>7678071

>>7678044
It can't but remember how much anger there was over fucking 1.5 loosely replicating shit like Greg Rutkowski? NAI's styling is considerably closer.

Anonymous

Anonymous Mon 27 Nov 2023 20:15:06 No.7678054 Report

Quoted By: >>7678071

>>7678036
Something like adetailer might be able to do it. When I have time, I've been trying to get it working with NAI, if it works the way I've been assuming it does, it should be doable, but I haven't had time to really look into it and could be wrong.

>>7678044
>Shading and color grading is usually pretty off and linework is inconsistent at best
Are you seriously implying lora's do any better?

Half the fun is mixing artists and creating weird hybrid styles anyway.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315364-4021567432-by b (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 20:16:02 No.7678056 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315379-4065249153-(by (...).png, 2MiB, 1024x1024

Anonymous Mon 27 Nov 2023 20:16:20 No.7678057 Report

Quoted By: >>7678068

Anonymous

Anonymous Mon 27 Nov 2023 20:17:08 No.7678059 Report

Quoted By: >>7678060 >>7678062 >>7678064

>>7677735
Just don't mention any artist's name directly. I certainly haven't had any trouble - though I don't try to make any cash.

The type of person who'd actually try to report you for copying a style probably has ai filtered anyway.

Pixiv barely even cares about censoring - they do the bare minimum they can get away with. There's a whole uncensored tag too.

Anonymous

Anonymous Mon 27 Nov 2023 20:17:45 No.7678060 Report

Quoted By: >>7678078

>>7678059
thanks, I'll try posting some stuff then :) makes my life easier. I'll also clean up metadata so that they can't complain about the artist tag

Anonymous

Anonymous Mon 27 Nov 2023 20:18:25 No.7678062 Report

Quoted By:

>>7678059
Damn seems like I cared too fucking much about this

Anonymous

Anonymous Mon 27 Nov 2023 20:19:16 No.7678064 Report

Quoted By:

>>7678059
and yeah, I don't do it for cash either, just for pixiv (You)'s

Anonymous

Anonymous Mon 27 Nov 2023 20:19:33 No.7678065 Report

Quoted By: >>7678072

I don't understand the posting AI art to pixiv phenomenon
I just wanna make my own gens, share them in a really small circle like an /h/ thread and then other than that they're just my personal use

Anonymous

Anonymous Mon 27 Nov 2023 20:22:50 No.7678068 Report

Quoted By: >>7678069

>>7678057
average kawakami rokkaku enjoyer
I like the tape over her legs. does he look a bit ultra lanky to you or is it just me.

Anonymous

Anonymous Mon 27 Nov 2023 20:23:46 No.7678069 Report

Quoted By:

>>7678068
yeah it picked up on the slenderness of the average character he draws and now they are all like 7ft tall

Anonymous

Anonymous Mon 27 Nov 2023 20:24:13 No.7678071 Report

Quoted By:

>>7678049
I'm already aware I'm a snob with dumb standards for art but I still don't see the point in wanting to replicate 1:1 unless you know literally just that one artist. I can understand an amount of it for novelty but doing things only while trying to directly replicate the direct style of someone else still just seems really boring.
>>7678052
Yeah, because westerners and literal retards hyper reacting to new things. NAI style copying only really impacts nips and it impacts, specifically, porn artists the most.
Even if there was some huge outrage, unless it hit the nipland mainstream/trending you wouldn't really hear about it without specifically looking for it.
>>7678054
>Are you seriously implying lora's do any better?
I never said that did I?
Both can be bad at trying to replicate an artist 1:1.
That said, lora's for styling are still honestly preferable, especially if you're the one making them yourself and can control what contributes to it, how much burn in it has, etc.

Anonymous

Anonymous Mon 27 Nov 2023 20:24:14 No.7678072 Report

Quoted By:

>>7678065
just (you)s
posting on /hdg/ would be cozy if there wasn't so much occasional schizophrenia here

Anonymous

Anonymous Mon 27 Nov 2023 20:28:12 No.7678077 Report

Quoted By:

>have even less reasons to build my new rig now

Anonymous

Anonymous Mon 27 Nov 2023 20:29:03 No.7678078 Report

Quoted By: >>7678079 >>7678083

>>7678060
I thnik Pixiv wipes metadata anyway, at least I've never managed to import an AI pic from there.

Anonymous

Anonymous Mon 27 Nov 2023 20:29:31 No.7678079 Report

Quoted By:

>>7678078
it doesn't. those were tards whove never looked at a settings menu in their life

Anonymous

View Same Google ImgOps iqdb SauceNAO b5d9c4ab-c40e-4d99-a5fe-7771ff06 (...).png, 2MiB, 1280x1856

Anonymous Mon 27 Nov 2023 20:30:53 No.7678081 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 20:31:29 No.7678083 Report

Quoted By:

>>7678078
>I thnik Pixiv wipes metadata anyway
nope. and with nai you gotta wipe the alpha as well

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_ehxywt.png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 20:33:15 No.7678086 Report

Quoted By: >>7678129

>>7677380
female pov is very difficult for nai. this is about the best I got and even then it's 10% chance of generating. Looking at danbooru, there are only like actual 20 images of female pov masturbation; there are a few more of mirror reflections but I don't think SDXL's anatomy knowledge is advanced enough for this yet. not without a lora of specifically this pose.

Anonymous

Anonymous Mon 27 Nov 2023 20:34:24 No.7678087 Report

Quoted By: >>7678091 >>7678132

Is NovelAI constantly just scraping Danbooru so even if art gets deleted it still gets put in

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315382-2579350764-(by (...).png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 20:34:56 No.7678088 Report

Quoted By: >>7678101

i really cant get enough of this artist

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315384-3198884818-(by (...).png, 1MiB, 1024x1024

Anonymous Mon 27 Nov 2023 20:35:13 No.7678090 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 20:35:15 No.7678091 Report

Quoted By:

>>7678087
>Is NovelAI constantly just scraping Danbooru so even if art gets deleted it still gets put in
check status:deleted on danbooru, but yeah, you're right, they are probably saving the datasets because some artist do takedown request to danbooru

Anonymous

Anonymous Mon 27 Nov 2023 20:36:43 No.7678093 Report

Quoted By:

>>7677347
My secret is to upscale and inpaint in local. Even the details. Better yet if you have such styles as loras. Also mix artists.

Anonymous

Anonymous Mon 27 Nov 2023 20:38:00 No.7678096 Report

Quoted By: >>7678099

>Control tools not supported for this model

I have Opus tier and it's my first time trying Img2Img. Why are tools disabled?

Anonymous

Anonymous Mon 27 Nov 2023 20:39:12 No.7678099 Report

Quoted By:

>>7678096
It's not ready yet. I think NAI said something along the lines that controlnet for V3 has to be trained and they didn't want to delay launching the model due to it.

Anonymous

Anonymous Mon 27 Nov 2023 20:39:48 No.7678101 Report

Quoted By: >>7678105

>>7678088
>he missed out on all the wholesome might-as-well-be-canon Holo doujins

Anonymous

Anonymous Mon 27 Nov 2023 20:42:28 No.7678105 Report

Quoted By: >>7678109

>>7678101
theres a finite number of them

Anonymous

Anonymous Mon 27 Nov 2023 20:43:38 No.7678109 Report

Quoted By:

>>7678105
still more than you'd expect

Anonymous

Anonymous Mon 27 Nov 2023 20:43:56 No.7678111 Report

Quoted By: >>7678114 >>7678115 >>7678116 >>7678118 >>7678135

Would I get in trouble for generating...you-know-whats?

Anonymous

Anonymous Mon 27 Nov 2023 20:44:41 No.7678114 Report

Quoted By: >>7678123

>>7678111
chicks with dicks?

Anonymous

Anonymous Mon 27 Nov 2023 20:44:45 No.7678115 Report

Quoted By:

>>7678111
not locally

Anonymous

Anonymous Mon 27 Nov 2023 20:45:04 No.7678116 Report

Quoted By:

>>7678111
no, I don't know.
explain further.
in great detail.

Anonymous

Anonymous Mon 27 Nov 2023 20:45:42 No.7678118 Report

Quoted By:

>>7678111
for the love of god this is the Nth time
anon just fucking say it we can't read minds

Anonymous

View Same Google ImgOps iqdb SauceNAO IMG_1596.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 20:46:28 No.7678119 Report

Quoted By: >>7678139 >>7678152

Any good artist tags for ass shaking, twerking, etc.? Fellatrix seems to be the only one consistently good so far.

Anonymous

Anonymous Mon 27 Nov 2023 20:47:11 No.7678122 Report

Quoted By:

>>7678008
Get someone who can to buy you a gift key.

Anonymous

Anonymous Mon 27 Nov 2023 20:47:35 No.7678123 Report

Quoted By: >>7678124

>>7678114
nta but how could that get you in trouble? or maybe til futas are illegal

Anonymous

Anonymous Mon 27 Nov 2023 20:48:41 No.7678124 Report

Quoted By:

>>7678123
it was a joke, they are /d/ even though the /d/ thread is nothing but 1girl futa and posting about delusional fanfics i still can't post them here

Anonymous

Anonymous Mon 27 Nov 2023 20:49:24 No.7678125 Report

Quoted By: >>7678128

>>7678036
In my experience doing img2img on the cropped face portion with a relevant prompt, which is all adetailer is, actually makes the result look worse with NAIv3. Or if it does look better then there's a style or detail level mismatch because of how close up it is.

Anonymous

Anonymous Mon 27 Nov 2023 20:50:55 No.7678128 Report

Quoted By: >>7678130 >>7678138

>>7678125
NAI's inpaint is kinda dogshit because it loses quality on inpaints. Someone said it was a VAE but no idea.

Anonymous

Anonymous Mon 27 Nov 2023 20:51:43 No.7678129 Report

Quoted By:

>>7678086
that is still reasonable good desu, I picked a pose I intentionally knew would have very little data

Anonymous

Anonymous Mon 27 Nov 2023 20:53:00 No.7678130 Report

Quoted By:

>>7678128
Well I wasn't even inpainting, just doing straight img2img on the face I cropped myself. Inpainting actually seems to work fine for me but I've only ever done one or two passes for minimal changes.

Anonymous

Anonymous Mon 27 Nov 2023 20:54:02 No.7678132 Report

Quoted By:

>>7678087
>constantly
Probably just before a new training run.

Anonymous

Anonymous Mon 27 Nov 2023 20:55:50 No.7678135 Report

Quoted By:

>>7678111
Generate all the scat you like, but it's not very good at it.

Anonymous

Anonymous Mon 27 Nov 2023 20:57:47 No.7678138 Report

Quoted By: >>7678144 >>7678148 >>7678151

>>7678128
When you inpaint without the "copy original image" box ticked, the whole image gets encoded by the VAE and then decoded by the VAE. The VAE is lossy, like JPEG, so every time you lose some quality. If you tick the box, you may get seams around your inpaint, but those can probably get fixed by a very low strength img2img/enhance at the end.

Anonymous

Anonymous Mon 27 Nov 2023 20:57:58 No.7678139 Report

Quoted By: >>7678142

>>7678119
Surprised you managed to get a non-cellulite ass from that fag

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, Jessica Rabbi (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 20:58:14 No.7678140 Report

Quoted By:

What's the optimal way to combine artists in NAI?
Both artists in succession? One at the start one at the end? One with emphasis and one without? Or just throw shit at the wall and see what sticks

Anonymous

View Same Google ImgOps iqdb SauceNAO IMG_1597.png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 20:59:40 No.7678142 Report

Quoted By: >>7678152 >>7678177

>>7678139
Maybe it’s because I’m tagging aqua? Either way, I found out Jinu kind of works too

Anonymous

Anonymous Mon 27 Nov 2023 21:00:10 No.7678144 Report

Quoted By: >>7678153

>>7678138
why is their vae so fucking bad though lol

Anonymous

View Same Google ImgOps iqdb SauceNAO 22228-1912008342.png, 2MiB, 1280x1600

Anonymous Mon 27 Nov 2023 21:00:58 No.7678145 Report

Quoted By: >>7678158 >>7678160 >>7678797

yes, i like img2img, controlnet, loras, upscaling, inpainting and aom merges.

Anonymous

Anonymous Mon 27 Nov 2023 21:01:16 No.7678148 Report

Quoted By: >>7678156

>>7678138
This is the correct way to do it. There ideally should be control in their UI to adjust mask feathering and whatnot but I always apply a very low strength img2img when I'm done anyway as you said.

Anonymous

Anonymous Mon 27 Nov 2023 21:03:42 No.7678151 Report

Quoted By: >>7678157 >>7678163 >>7678165

>>7678138
>very low strength img2img/enhance at the end.
But... then the image goes through the VAE anyway? It doesn't make sense to me

Anonymous

Anonymous Mon 27 Nov 2023 21:03:57 No.7678152 Report

Quoted By:

>>7678142
>>7678119
AQUA ASSJOB

Anonymous

Anonymous Mon 27 Nov 2023 21:05:01 No.7678153 Report

Quoted By:

>>7678144
it's always like that
You can just run encode-decode on comfy, or do and img2img pass at 0.01 denoise to see the difference. Usually a little bit of detail loss and hue shifting shadows towards purple.

Anonymous

Anonymous Mon 27 Nov 2023 21:08:53 No.7678156 Report

Quoted By:

>>7678148
i'm gonna do it

i'm gonna manual tag

Anonymous

Anonymous Mon 27 Nov 2023 21:09:04 No.7678157 Report

Quoted By: >>7678170

>>7678151
nta but I assume it's copying the latent of the original image to use for inpaint, so it stays VAE encoded from the time you first generated it until it is complete. As opposed to getting VAE decoded into .png then re-encoded for inpainting.

Would be a lot easier to explain with comfy nodes.

Anonymous

Anonymous Mon 27 Nov 2023 21:09:31 No.7678158 Report

Quoted By:

>>7678145
>i am not coping, i am based because i img2img'd a meme

Anonymous

Anonymous Mon 27 Nov 2023 21:10:48 No.7678160 Report

Quoted By:

>>7678145
That's fine. Not hentai though.

Anonymous

Anonymous Mon 27 Nov 2023 21:12:23 No.7678163 Report

Quoted By:

>>7678151
But only once, which is fine.

Anonymous

Anonymous Mon 27 Nov 2023 21:13:18 No.7678165 Report

Quoted By:

>>7678151
only once and you're slightly re-noising and denoising the entire image as a whole

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{buttjob}}}, 1girl, 1boy, peni (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 21:17:41 No.7678169 Report

Quoted By: >>7678176 >>7679951

Anonymous

View Same Google ImgOps iqdb SauceNAO 123.jpg, 246KiB, 1920x1080

Anonymous Mon 27 Nov 2023 21:18:18 No.7678170 Report

Quoted By:

>>7678157
simplified view, just follow the purple lines

Anonymous

Anonymous Mon 27 Nov 2023 21:23:51 No.7678176 Report

Quoted By: >>7678179 >>7678181 >>7678806

>>7678169
Where did you get this photo of me and Aqua. That's me cumming on Aqua's butt

Anonymous

View Same Google ImgOps iqdb SauceNAO 1boy, 1girl, {{{metera (granblue (...).png, 3MiB, 1280x1856

Anonymous Mon 27 Nov 2023 21:23:52 No.7678177 Report

Quoted By:

>>7678142
Artist?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00019-501588963.png, 2MiB, 1024x1536

Anonymous Mon 27 Nov 2023 21:24:29 No.7678178 Report

Quoted By: >>7678185 >>7678229

>My face when additional networks is broken now
I miss it. It was better, it was, I dare say, king.

Anonymous

Anonymous Mon 27 Nov 2023 21:24:39 No.7678179 Report

Quoted By:

>>7678176
are you an anime person

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{buttjob}}}, 1girl, 1boy, peni (...).png, 912KiB, 832x1216

Anonymous Mon 27 Nov 2023 21:25:43 No.7678181 Report

Quoted By: >>7678188 >>7679951

>>7678176
Good thing that this one isn't (You) anon

Anonymous

View Same Google ImgOps iqdb SauceNAO 1689740262976763.png, 41KiB, 158x121

Anonymous Mon 27 Nov 2023 21:28:52 No.7678185 Report

Quoted By: >>7678345

>>7678178
the fuck is this

Anonymous

Anonymous Mon 27 Nov 2023 21:29:13 No.7678186 Report

Quoted By: >>7678197

can it do stuff like anal gaping and cum vomiting?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00874-1885385870.png, 1MiB, 832x1152

Anonymous Mon 27 Nov 2023 21:32:24 No.7678187 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 21:33:06 No.7678188 Report

Quoted By: >>7678189

>>7678181
HOLYYYYYYYYYYYYYYYYYYY
(Btw is there a tag to get the cock in between, sandwiched all the way in the cheeks?)

Anonymous

Anonymous Mon 27 Nov 2023 21:33:46 No.7678189 Report

Quoted By:

>>7678188
hotdogging?

Anonymous

View Same Google ImgOps iqdb SauceNAO tsubasa tsubasa, koseki bijou, n (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 21:37:22 No.7678192 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO saten ruiko, toaru kagaku no rai (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 21:39:58 No.7678197 Report

Quoted By: >>7678198 >>7678201

>>7678186
kinda can do vomiting cum

Anonymous

Anonymous Mon 27 Nov 2023 21:40:18 No.7678198 Report

Quoted By: >>7678200 >>7678809

>>7678197
why do you gen censored

Anonymous

View Same Google ImgOps iqdb SauceNAO ComfyUI_03706_.png, 384KiB, 640x640

Anonymous Mon 27 Nov 2023 21:40:50 No.7678199 Report

Quoted By: >>7678215

Is there a good neg prompts for shiny skin besides "shiny skin, shiny"?

Anonymous

Anonymous Mon 27 Nov 2023 21:41:04 No.7678200 Report

Quoted By: >>7678204 >>7678208

>>7678198
forgot to put uncensored tag

Anonymous

Anonymous Mon 27 Nov 2023 21:41:07 No.7678201 Report

Quoted By: >>7678211

>>7678197
neat, thanks
og taimanin asagi ruined me

Anonymous

Anonymous Mon 27 Nov 2023 21:43:02 No.7678204 Report

Quoted By: >>7678216

>>7678200
it always gens uncensored for me by default, do you perhaps copy tags from danbooru and don't remove censored? the userscript I have (modified version of someone else's) just ignores that and also appends stuff like artist: to artist tags (copyright: and character: for those as well)

Anonymous

Anonymous Mon 27 Nov 2023 21:43:52 No.7678208 Report

Quoted By: >>7678216

>>7678200
you can just put censored in the negatives

Anonymous

Anonymous Mon 27 Nov 2023 21:45:02 No.7678211 Report

Quoted By: >>7678810

>>7678201
>it can do sakura
be right back breaking my dick

Anonymous

Anonymous Mon 27 Nov 2023 21:45:51 No.7678215 Report

Quoted By: >>7678219

>>7678199
As someone who has actually had sex irl, buttcheeks are never that shiny.

Anonymous

Anonymous Mon 27 Nov 2023 21:45:52 No.7678216 Report

Quoted By: >>7678231

>>7678204
>do you perhaps copy tags from danbooru and don't remove censored
nope there's no "censored" tag in there
>it always gens uncensored for me by default
I think it depends on the artist prompts
>>7678208
yeah I use both when I need, I just wanted to do a quick test for anon.

Anonymous

Anonymous Mon 27 Nov 2023 21:46:53 No.7678219 Report

Quoted By:

>>7678215
BOOOOOOO FAKECEL

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, taimanin yukikaze, taiman (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 21:49:29 No.7678228 Report

Quoted By: >>7678230

uhh?

Anonymous

View Same Google ImgOps iqdb SauceNAO 22203-4086470136.png, 3MiB, 1280x1600

Anonymous Mon 27 Nov 2023 21:49:51 No.7678229 Report

Quoted By: >>7678277 >>7678345

>>7678178
yup. you can still use it if you restart the webui server when switching between txt2img and img2img. or use it just for checking tags lol.

Anonymous

Anonymous Mon 27 Nov 2023 21:51:14 No.7678230 Report

Quoted By: >>7678238

>>7678228
wrong tag my man, it's mizuki yukikaze with taimanin yukikaze and taimanin (series) being series tags

Anonymous

Anonymous Mon 27 Nov 2023 21:51:59 No.7678231 Report

Quoted By: >>7678235

>>7678216
Some artists like mizumizuni make it impossible to gen uncensored even with the tag. The only way is to remove the artist tag and inpaint the censors.

Anonymous

Anonymous Mon 27 Nov 2023 21:53:15 No.7678235 Report

Quoted By: >>7678243

>>7678231
>inpaint the censors
Is this hard to do?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1girl, mizuki yukikaze, taimanin (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 21:53:53 No.7678238 Report

Quoted By: >>7678244

>>7678230
Oh thanks, I actually don't really know the series but liked this girl lol.

Anonymous

Anonymous Mon 27 Nov 2023 21:55:36 No.7678243 Report

Quoted By:

>>7678235
Remove artist tag, add censored to negative, uncensored to prompt, click the inpaint button at the top and roughly paint a mask over the censored areas, then generate. It's pretty easy.

Anonymous

Anonymous Mon 27 Nov 2023 21:55:36 No.7678244 Report

Quoted By:

>>7678238
she's cute
the series went to shit after they fired the absolute chad who animated the first 4 eps

Anonymous

Anonymous Mon 27 Nov 2023 22:05:39 No.7678256 Report

Quoted By: >>7678257 >>7678260 >>7678270 >>7678282

I really wish they'd train on other boorus to get even more characters but:
Gelbooru is mostly just a Danbooru mirror
Sankaku is getting ultra-spammed with furry porn after the search engine changes and AI shitshow plus the tagging work on there is objectively shit
Yandere doesn't have a lot of stuff but has quite a few paid posts that you'd find on Kemono
R34 (both sites) are mostly a lost cause

We should seriously just create our own booru but the tagging quality isn't even the entire problem, it's the fact that tagging itself sucks and we really need branching tags.

Anonymous

Anonymous Mon 27 Nov 2023 22:06:31 No.7678257 Report

Quoted By: >>7678261

>>7678256
they just should doing synthetic tagging which will solve most problems

Anonymous

Anonymous Mon 27 Nov 2023 22:07:19 No.7678260 Report

Quoted By: >>7678264 >>7678271

>>7678256
Do deleted Danbooru artworks still get scraped and therefore NovelAI still gets them?

Anonymous

Anonymous Mon 27 Nov 2023 22:07:50 No.7678261 Report

Quoted By:

>>7678257
You mean auto/ai tagging?
Sure but what do you think the autotagger will be trained on? Even the furfaggots don't have good tagging, people think they do just because they see mile long tag lists on e6.

Anonymous

Anonymous Mon 27 Nov 2023 22:08:57 No.7678264 Report

Quoted By:

>>7678260
I have no idea, sorry.

Anonymous

Anonymous Mon 27 Nov 2023 22:12:51 No.7678270 Report

Quoted By: >>7678278

>>7678256
What exactly is branching tags?

Anonymous

Anonymous Mon 27 Nov 2023 22:13:34 No.7678271 Report

Quoted By: >>7678276

>>7678260
m8
get it through your head already
there is not a single thing you can personally do to change, influence or improve anything NAI does
and you shouldn't fucking want to, either. they're closed source and have a paywall.
They do not give a fuck about you or your ideas beyond your wallet.

Anonymous

Anonymous Mon 27 Nov 2023 22:14:39 No.7678276 Report

Quoted By: >>7678279 >>7678285

>>7678271
le company bad

Anonymous

Anonymous Mon 27 Nov 2023 22:15:13 No.7678277 Report

Quoted By:

>>7678229
i use it for embed comments and shit in loras

Anonymous

Anonymous Mon 27 Nov 2023 22:16:30 No.7678278 Report

Quoted By: >>7678281

>>7678270
Tags are associated with each character or subject in general instead of being just a list, sorta like how you use BREAK on local when you do multiple characters or want tokens to get more attention by prompting in blocks

Anonymous

Anonymous Mon 27 Nov 2023 22:17:41 No.7678279 Report

Quoted By: >>7678280

>>7678276
yes, you retard

Anonymous

Anonymous Mon 27 Nov 2023 22:18:38 No.7678280 Report

Quoted By: >>7678284

>>7678279
go back to making waifu diffusion retard

Anonymous

Anonymous Mon 27 Nov 2023 22:19:22 No.7678281 Report

Quoted By:

>>7678278
I see, thanks.

Anonymous

Anonymous Mon 27 Nov 2023 22:19:32 No.7678282 Report

Quoted By: >>7678285 >>7678286

>>7678256
imagine an exhentai finetune... would be fucking hell to tag sadly

Anonymous

Anonymous Mon 27 Nov 2023 22:19:34 No.7678284 Report

Quoted By:

>>7678280
such a retarded loser

Anonymous

Anonymous Mon 27 Nov 2023 22:21:07 No.7678285 Report

Quoted By: >>7678293

>>7678276
unironically, yes.
and you'd have to be clinically retarded to want to try and """"help"""" a paid service FOR FREE while still giving them fucking money for it.
>>7678282
Would be literal garbage.
It's like you people have absolutely no idea what even makes an image "good" or "bad" for a fucking dataset.

Anonymous

Anonymous Mon 27 Nov 2023 22:21:45 No.7678286 Report

Quoted By:

>>7678282
Exhentai has a ton of shit too and they do remove stuff. Personally I think that keeping VN CGs in the dataset is just asking for trouble

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315396-3966440485-kawa (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 22:24:24 No.7678289 Report

Quoted By:

Anonymous

Anonymous Mon 27 Nov 2023 22:26:23 No.7678293 Report

Quoted By: >>7678297 >>7678300 >>7678301

>>7678285
>unironically, yes.
>and you'd have to be clinically retarded to want to try and """"help"""" a paid service FOR FREE while still giving them fucking money for it.

I adore how mad you are. I'll do it and I'll do it for free, I want it to get better and better just so that people like you can keep seething. If this sounds petty to you imagine how petty you sound when you're malding.

Don't forget the monthly payment for the 4090 you bought just to coom to AOM slop.

Anonymous

Anonymous Mon 27 Nov 2023 22:28:43 No.7678297 Report

Quoted By:

>>7678293
>coom to AOM slop
nobody does this for like half a year

Anonymous

Anonymous Mon 27 Nov 2023 22:30:35 No.7678300 Report

Quoted By:

>>7678293
>I'll do it and I'll do it for free
I don't know how considering there is absolutely nothing you can do. And I bet if you went to their discord and begged and pleaded they'd tell you "thanks, no thanks, we want literally nothing to do with you, fuck off" just the same.
And you should really just fuck off and actually READ NIGGA READ and learn how this shit works because you clearly have zero understanding of it.

Anonymous

Anonymous Mon 27 Nov 2023 22:30:51 No.7678301 Report

Quoted By:

>>7678293
You aren't doing anything you fucking retard. You're just making a fool out of yourself.

Anonymous

Anonymous Mon 27 Nov 2023 22:31:58 No.7678303 Report

Quoted By:

:^)

Anonymous

Anonymous Mon 27 Nov 2023 22:38:54 No.7678312 Report

Quoted By: >>7678314 >>7678317 >>7678324

Becoming a tag gardener on Danbooru is a good thing. It'll help NAI and it'll also help open models trained on Danbooru and it make Danbooru itself better as well. If you have the necessary time and autism to upload images there and tag comprehensively (a lot of images are badly undertagged), by all means do it.

Anonymous

Anonymous Mon 27 Nov 2023 22:39:30 No.7678314 Report

Quoted By:

>>7678312
shit website

Anonymous

Anonymous Mon 27 Nov 2023 22:41:17 No.7678317 Report

Quoted By: >>7678318 >>7678321

>>7678312
complete waste of effort because AI will perfectly tag images soon enough

Anonymous

Anonymous Mon 27 Nov 2023 22:41:56 No.7678318 Report

Quoted By: >>7678322

>>7678317
And what are those perfect taggers are going to be trained on?

Anonymous

Anonymous Mon 27 Nov 2023 22:42:45 No.7678321 Report

Quoted By:

>>7678317
you reap what you sow
shitty tagging in shitty ai tagging out

Anonymous

Anonymous Mon 27 Nov 2023 22:43:36 No.7678322 Report

Quoted By:

>>7678318
real life

Anonymous

Anonymous Mon 27 Nov 2023 22:44:29 No.7678324 Report

Quoted By: >>7678325 >>7678326

>>7678312
pretty sure any pre-existing tags get completely fucking discarded, for good reason(mistags, tag concatenation, dumb meme tags, etc.)
Every large dataset gets autotagged in some manner that's consistent. The issue with tagging is that current autotaggers are mediocre/trash and even going to longer route of setting up a machine vision llm tagger for natural language, the models are also largely trash and not super great. Supposedly the current "best" (local) llm model for that is one that was reverse trained off of GPT-4 outputs, which is all sorts of fucked.

Anonymous

Anonymous Mon 27 Nov 2023 22:45:28 No.7678325 Report

Quoted By:

>>7678324
damn this is fucking depressing to read

Anonymous

Anonymous Mon 27 Nov 2023 22:45:55 No.7678326 Report

Quoted By: >>7678328

>>7678324
>pretty sure any pre-existing tags get completely fucking discarded, for good reason(mistags, tag concatenation, dumb meme tags, etc.)
Pretty sure you're completely fucking wrong.

Anonymous

Anonymous Mon 27 Nov 2023 22:48:21 No.7678328 Report

Quoted By: >>7678332

>>7678326
in NAI's case they probably leave some amount of metadata like artist/date/character/IP, which are their own separate and super specific tags(and generally don't need "help" to get tagged).
But descriptors? There's no reason to keep them.

Anonymous

Anonymous Mon 27 Nov 2023 22:50:44 No.7678332 Report

Quoted By: >>7678336 >>7678351 >>7678366

>>7678328
Human made Danbooru tags are still on the level where they're more accurate than the best autotaggers. It's not like there's a better dataset to train such a tagger on after all. If you add data from somewhere else? Sure, you'll put it in the Danbooru auto tagger, but you'd have to be retarded to throw away the best labels you can get.

Anonymous

Anonymous Mon 27 Nov 2023 22:53:05 No.7678336 Report

Quoted By:

>>7678332
>Human made Danbooru tags are still on the level where they're more accurate than the best autotaggers.
lol

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, {sonia (Fire (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 22:54:39 No.7678340 Report

Quoted By: >>7678369

Combining Mimonel and Cle Masahiro, surprisingly not awful

Anonymous

View Same Google ImgOps iqdb SauceNAO artist_cutesexyrobutts, izayoi a (...).png, 1010KiB, 832x1216

Anonymous Mon 27 Nov 2023 22:56:06 No.7678341 Report

Quoted By:

imagine the smell

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, very aesthetic, [[cutesexy (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 22:58:17 No.7678345 Report

Quoted By: >>7678348 >>7678356

>>7678185
Who knows? but her expression was funny.
>>7678229
Any way to use it with loras in the lora folder? I don't wanna double my lora storage use by copying them.

Anonymous

Anonymous Mon 27 Nov 2023 22:59:21 No.7678348 Report

Quoted By: >>7678349 >>7678350

>>7678345
Where the fuck is that dick coming from?

Anonymous

Anonymous Mon 27 Nov 2023 23:00:01 No.7678349 Report

Quoted By:

>>7678348
impaled through her neck

Anonymous

Anonymous Mon 27 Nov 2023 23:00:09 No.7678350 Report

Quoted By: >>7678353

>>7678348
His body

Anonymous

Anonymous Mon 27 Nov 2023 23:00:10 No.7678351 Report

Quoted By: >>7678352

>>7678332
>Human made Danbooru tags are still on the level where they're more accurate than the best autotaggers
lmao
Some of them? Sure. All of them? No. Not even fucking close. There are a lot of garbage irrelevant tags for the same fucking thing(but under a different tag because uh reasons), a lot of mistagging and just inconsistency all around.
And, again, this is straight up about consistency. Any amount of inconsistency with manual tags taints the whole thing. It's better to have a subpar but consistent autotag pool than it is to have an inconsistent mix.

Anonymous

Anonymous Mon 27 Nov 2023 23:01:30 No.7678352 Report

Quoted By:

>>7678351
also a lot of posts don't have the same level of tagging detail as the most popular ones

Anonymous

Anonymous Mon 27 Nov 2023 23:01:43 No.7678353 Report

Quoted By:

>>7678350
you know what I mean

Anonymous

Anonymous Mon 27 Nov 2023 23:05:12 No.7678356 Report

Quoted By:

>>7678345
you can add extra paths for it in settings -> additional networks

Anonymous

Anonymous Mon 27 Nov 2023 23:06:09 No.7678357 Report

Quoted By: >>7678411 >>7678522

>>7676500
This is amazing. Could you make it so batch size/count works with inpainting as well? I get an indexerror when I try it.

Anonymous

Anonymous Mon 27 Nov 2023 23:13:27 No.7678366 Report

Quoted By:

>>7678332
>Human made Danbooru tags are still on the level where they're more accurate than the best autotaggers
yeah for the 0.2% that are done by someone with a brain

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, {sonia (Fire (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 23:15:25 No.7678369 Report

Quoted By:

>>7678340
Combining Mimonel and CSR, surprisingly not awful as well
it looks like putting Mimonel in front favors his shading/coloring so if you're getting an artist that's particularly blurry maybe try it out

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{anila (granblue fantasy)}}}, (...).png, 3MiB, 1856x1280

Anonymous Mon 27 Nov 2023 23:15:48 No.7678371 Report

Quoted By:

NovelaiGAWDs

Anonymous

Anonymous Mon 27 Nov 2023 23:18:23 No.7678375 Report

Quoted By: >>7678378 >>7678382 >>7678383

Wait Anlas points runs out really fucking quick

What the fuck 10,000 wasn't enough

Anonymous

Anonymous Mon 27 Nov 2023 23:19:57 No.7678378 Report

Quoted By: >>7678387

>>7678375
This is why you sub and never buy anals

Anonymous

View Same Google ImgOps iqdb SauceNAO K5.png, 500KiB, 512x768

Anonymous Mon 27 Nov 2023 23:20:13 No.7678379 Report

Quoted By: >>7679775

>>7676611
>>7676667
>>7676910
Thanks anon, appreciate it. Can't seem to get enough wholesome content of her, always seems to be way out of proportion or just way too intense for me.

>>7677253
Sure, here - I don't remember if I did any img2img though, might be a bit off: https://files.catbox.moe/sj9ucs.png
Toss me one for your image as well? Came out well.

Anonymous

Anonymous Mon 27 Nov 2023 23:20:52 No.7678382 Report

Quoted By: >>7678387

>>7678375
That is why you get Opus.

Anonymous

Anonymous Mon 27 Nov 2023 23:21:06 No.7678383 Report

Quoted By: >>7678386 >>7678387

>>7678375
How did you burn through it all so fast? I still have 7.9k and like at least 400 of that was caused by misclicks

Anonymous

Anonymous Mon 27 Nov 2023 23:22:19 No.7678386 Report

Quoted By:

>>7678383
maybe he was only doing higher res gens

also, fun fact, image gen is basically unusable for anyone except opus, imagine getting a $15 or $10 plan to get 1000 anlas that you can only gen 50 normal sized v3 images.

yes, a normal v3 gen costs 20 fucking anlas because it's still a high res gen, opus subs like me get it for free

Anonymous

Anonymous Mon 27 Nov 2023 23:22:34 No.7678387 Report

Quoted By: >>7678389 >>7678390

>>7678382
>>7678378
>>7678383
I did get Opus tho and I'm still running out of Anlas. Is it because of the size of my images or how many tags I'm putting in each time? I'm doing image2image

Anonymous

Anonymous Mon 27 Nov 2023 23:23:03 No.7678389 Report

Quoted By: >>7678392

>>7678387
if you're doing img2img and the selected res is over the default normal you'll be paying anlas for each gen

Anonymous

Anonymous Mon 27 Nov 2023 23:23:39 No.7678390 Report

Quoted By: >>7678392

>>7678387
Image2image costs anlas if the base image exceeds the 1024x1024 resolution parameters of the 3 default gen types

Anonymous

Anonymous Mon 27 Nov 2023 23:24:40 No.7678391 Report

Quoted By: >>7678705

Is the NAI IMG2IMG even good? I haven't felt any need to use it because the prompter is so good I usually get the exact composition I want fast

Anonymous

Anonymous Mon 27 Nov 2023 23:25:44 No.7678392 Report

Quoted By: >>7678396 >>7678398

>>7678389
>>7678390
Shiiiiiet

Anonymous

Anonymous Mon 27 Nov 2023 23:27:33 No.7678396 Report

Quoted By: >>7678402

>>7678392
There's a really easy way to tell if your current gen will cost you anals. Does the number on the generate button say "0"? If not, it'll cost you anals.

Anonymous

Anonymous Mon 27 Nov 2023 23:27:53 No.7678398 Report

Quoted By: >>7678402

>>7678392
If you got opus you can still gen infinitely at the default parameters (or lower but why would you want to) you just won't be able to upscale the good shit you get

Anonymous

Anonymous Mon 27 Nov 2023 23:29:37 No.7678402 Report

Quoted By: >>7678407

>>7678396
>>7678398
Yeah I was making huge images. Now I changed the setting to just Landscape mode (1216 x 832) And now I'm not losing Anlans points.

Anonymous

Anonymous Mon 27 Nov 2023 23:30:56 No.7678407 Report

Quoted By: >>7678413

>>7678402
>huge images
post an example, how huge are we talking

Anonymous

Anonymous Mon 27 Nov 2023 23:32:21 No.7678411 Report

Quoted By: >>7678427

>>7678357
I'll take a look at that once I've finished doing what I can to get adetailer working.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1boy, 1girl, kaname madoka (...).png, 1MiB, 2048x896

Anonymous Mon 27 Nov 2023 23:34:43 No.7678413 Report

Quoted By: >>7678417 >>7678419

>>7678407
All that just to not even get prone bone. Can you please help me to get prone bone position right? What's wrong with the tags I put?

Anonymous

Anonymous Mon 27 Nov 2023 23:38:23 No.7678417 Report

Quoted By:

>>7678413
Try using some combination of "lying on stomach" for the girl

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, {sonia (Fire (...).png, 1MiB, 1216x832

Anonymous Mon 27 Nov 2023 23:39:42 No.7678419 Report

Quoted By: >>7678423

>>7678413
This worked for me with 100% consistency so far in 10 gens

Anonymous

Anonymous Mon 27 Nov 2023 23:43:13 No.7678423 Report

Quoted By: >>7678429

>>7678419
Buttocks WAY too big for my tastes

Anonymous

Anonymous Mon 27 Nov 2023 23:44:44 No.7678427 Report

Quoted By: >>7678522

>>7678411
Based. Having batch size for nai feels like cheating.

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, ike (fire emblem), (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 23:46:10 No.7678428 Report

Quoted By:

I didn't expect it to do this well with either the artist or the character. The artist did draw a lot of female Ike though.

Anonymous

Anonymous Mon 27 Nov 2023 23:46:17 No.7678429 Report

Quoted By:

>>7678423
Well I have the huge ass tag on so

Anonymous

Anonymous Mon 27 Nov 2023 23:53:15 No.7678438 Report

Quoted By: >>7678439 >>7678457

>>7676500
does the inpainting for v3 suck like im being led to believe

Anonymous

Anonymous Mon 27 Nov 2023 23:55:11 No.7678439 Report

Quoted By: >>7678451 >>7678454

>>7678438
Yes. It is missing a ton of features from local, biggest one being no denoise setting so it completely changes the image every time. Don't see why img2img has denoise but not for inpaint.

Anonymous

View Same Google ImgOps iqdb SauceNAO anna (fire emblem), fire emblem (...).png, 1MiB, 832x1216

Anonymous Mon 27 Nov 2023 23:56:27 No.7678441 Report

Quoted By: >>7678479

Anonymous

Anonymous Tue 28 Nov 2023 00:01:21 No.7678451 Report

Quoted By:

>>7678439
do a cool custom inpaint by frankensteining together the good pieces of your img2img gens

Anonymous

Anonymous Tue 28 Nov 2023 00:02:58 No.7678454 Report

Quoted By: >>7678461

>>7678439
>biggest one being no denoise setting so it completely changes the image every time.
You can do this with the overlay feature but it causes seams in some cases

Anonymous

Anonymous Tue 28 Nov 2023 00:04:50 No.7678457 Report

Quoted By:

>>7678438
Not entirely. >>>/e/2821405
But it's certainly missing features compared to local inpaint, so if you want to fix eyes, you have to manually cut out the face, scale it up, put it back in, then copy it back and all that.

Anonymous

Anonymous Tue 28 Nov 2023 00:07:09 No.7678461 Report

Quoted By: >>7678467

>>7678454
Yah but you still can't set the actual denoise.

Anonymous

Anonymous Tue 28 Nov 2023 00:08:43 No.7678467 Report

Quoted By:

>>7678461
it doesn't warp the entire image though like having the overlay off

Anonymous

View Same Google ImgOps iqdb SauceNAO b4741ba6-4227-479b-ab5f-dadfb5ca (...).png, 638KiB, 1216x832

Anonymous Tue 28 Nov 2023 00:12:03 No.7678472 Report

Quoted By: >>7678473

Any tags to prevent/decrease body deformations like this? img2img btw

Anonymous

Anonymous Tue 28 Nov 2023 00:14:47 No.7678473 Report

Quoted By:

>>7678472
desu img2img without controlnet for sex poses is probably futile

Anonymous

Anonymous Tue 28 Nov 2023 00:17:58 No.7678479 Report

Quoted By: >>7678525

>>7678441
UAAAAAAAA, YOU BETTER CENSOR THIS!!!!!!

Anonymous

Anonymous Tue 28 Nov 2023 00:29:13 No.7678488 Report

Quoted By: >>7678493 >>7678503 >>7678511 >>7678615 >>7678815 >>7679187

NAIchads which samplers are you using? I was testing all of them but then I got bored and just stuck it on SDE. 2M Exponential is also good and so is regular non-ancestral Euler. Karras scheduler seems oddly broken on all samplers but it creates very soft, very painting-esque images with SDE when it doesn't fuck up the anatomy.

Anonymous

Anonymous Tue 28 Nov 2023 00:34:41 No.7678493 Report

Quoted By:

>>7678488
default euler

Anonymous

View Same Google ImgOps iqdb SauceNAO 1680440761168925.png, 2MiB, 1024x1024

Anonymous Tue 28 Nov 2023 00:38:47 No.7678503 Report

Quoted By:

>>7678488
euler

Anonymous

Anonymous Tue 28 Nov 2023 00:40:27 No.7678506 Report

Quoted By:

are there any alternate uis for NAI? stuff that uses the api
i know about the webui extension but i don't want to use gradio

Anonymous

View Same Google ImgOps iqdb SauceNAO 2girls, deeply affectionate, {{s (...).png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 00:41:59 No.7678511 Report

Quoted By:

Dick is fucking breaking lads, yuri can now be just raw prompted with some great results.

>>7678488
SDE seems to work for me. Honestly it really doesn't matter as a few of my better outputs were on euler.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{anila (granblue fantasy)}}}, (...).png, 3MiB, 1856x1280

Anonymous Tue 28 Nov 2023 00:42:53 No.7678512 Report

Quoted By:

Anonymous

Anonymous Tue 28 Nov 2023 00:45:24 No.7678522 Report

Quoted By: >>7678585

>>7678357
>>7678427
Batching in Image2Image should be working now.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{elise (fire emblem)}}}, {{{fi (...).png, 3MiB, 1280x1856

Anonymous Tue 28 Nov 2023 00:45:50 No.7678524 Report

Quoted By: >>7678606

Anonymous

View Same Google ImgOps iqdb SauceNAO anna (fire emblem), fire emblem (...).png, 1MiB, 1216x832

Anonymous Tue 28 Nov 2023 00:46:10 No.7678525 Report

Quoted By: >>7678606

>>7678479
Anna is too cute to censor

Anonymous

Anonymous Tue 28 Nov 2023 00:46:52 No.7678528 Report

Quoted By:

what's an artist with really good skin detail? not really looking for photorealism but i'm sure you get what i mean

Anonymous

Anonymous Tue 28 Nov 2023 00:51:50 No.7678543 Report

Quoted By: >>7678545 >>7678546

Can't believe it NAIv3 has made non ancestral euler cool again Does that mean its not the sampler fault but the model fault if it can't make all the samplers look good?

Anonymous

Anonymous Tue 28 Nov 2023 00:53:16 No.7678545 Report

Quoted By:

>>7678543
I really don't think there's a correlation

Anonymous

Anonymous Tue 28 Nov 2023 00:53:39 No.7678546 Report

Quoted By: >>7678560

>>7678543
>Can't believe it NAIv3 has made non ancestral euler cool again
No, anon, that's just you being retarded.
Euler was always good.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{elise (fire emblem)}}}, {{{fi (...).png, 3MiB, 1280x1856

Anonymous Tue 28 Nov 2023 00:58:23 No.7678558 Report

Quoted By: >>7678606

Anonymous

Anonymous Tue 28 Nov 2023 00:58:44 No.7678560 Report

Quoted By: >>7678568 >>7678570 >>7679756

>>7678546
If Euler was always good why did we abandoned it for DPM++ 2M Karras

Anonymous

Anonymous Tue 28 Nov 2023 01:00:18 No.7678568 Report

Quoted By: >>7678584

>>7678560
I think at the time that when we got Karras that it only needed like 20 steps to get results, and it was faster than Euler.

Anonymous

Anonymous Tue 28 Nov 2023 01:01:28 No.7678570 Report

Quoted By: >>7678572 >>7678592 >>7678595

>>7678560
>we
speak for yourself.
I haven't a/b tested samplers since UniPC came out and I've never wanted to swap off of Euler.

Anonymous

Anonymous Tue 28 Nov 2023 01:02:29 No.7678572 Report

Quoted By: >>7678586

>>7678570
>UniPC
no way people actually use that

Anonymous

Anonymous Tue 28 Nov 2023 01:06:09 No.7678584 Report

Quoted By:

>>7678568
Just because you're a slack brain too stupid to try anything new doesn't mean anyone else is.

Anonymous

Anonymous Tue 28 Nov 2023 01:06:21 No.7678585 Report

Quoted By: >>7678822 >>7679008

>>7678522
Based as fuck it works flawlessly. Another thing that would be awesome is the inpaint area function working with nai. Not sure if it's possible but would make fixing faces and stuff a breeze instead of having to go through the whole >crop in photoshop >inpaint in nai >splice back into image

Anonymous

Anonymous Tue 28 Nov 2023 01:06:40 No.7678586 Report

Quoted By:

>>7678572
iirc uniPC wasn't completely awful and actually (usually) tended to resemble euler a lot of the time, it's just that it favored SPEEEEEEEED so it wasn't always the best.
but I mean, that's just the last time I remember actually testing samplers against eachother. I was using euler before that and, clearly, testing it then didn't change anything since I'm still using euler now.

Anonymous

View Same Google ImgOps iqdb SauceNAO ComfyUI_temp_dpzbi_00238_.png, 1MiB, 1024x1024

Anonymous Tue 28 Nov 2023 01:07:44 No.7678592 Report

Quoted By: >>7678613

>>7678570
LCM or your wasting your time for no reason

Anonymous

Anonymous Tue 28 Nov 2023 01:08:34 No.7678595 Report

Quoted By:

>>7678570
Based, I used euler since February, never liked others every time I tried which was a lot

Anonymous

View Same Google ImgOps iqdb SauceNAO Marie1.jpg, 3MiB, 1695x2035

Anonymous Tue 28 Nov 2023 01:10:04 No.7678602 Report

Quoted By: >>7678610 >>7678630 >>7678652 >>7678665 >>7678686 >>7678804 >>7679660 >>7679708

im losing motivation bros

Anonymous

View Same Google ImgOps iqdb SauceNAO AlearAI.jpg, 305KiB, 1536x2304

Anonymous Tue 28 Nov 2023 01:11:30 No.7678606 Report

Quoted By: >>7678621

>>7678524
>>7678525
>>7678558
Azura or Alear (F) please.

Anonymous

Anonymous Tue 28 Nov 2023 01:12:15 No.7678610 Report

Quoted By: >>7678617

>>7678602
honestly that's a bit too perfect, going with local on this but i wouldn't be surprised if it's inpainted naiv3

Anonymous

Anonymous Tue 28 Nov 2023 01:14:04 No.7678613 Report

Quoted By: >>7678661

>>7678592
I've heard it needs a "kohya fix" is that true?

Anonymous

View Same Google ImgOps iqdb SauceNAO artist_uechin ewokaku, [[artist_ (...).png, 2MiB, 1280x1856

Anonymous Tue 28 Nov 2023 01:14:19 No.7678615 Report

Quoted By: >>7678618

>>7678488
Gen with DDIM, upscale with DPM++ 2M & SMEA

Anonymous

Anonymous Tue 28 Nov 2023 01:14:53 No.7678617 Report

Quoted By: >>7678629

>>7678610
thanks, it's local

Anonymous

Anonymous Tue 28 Nov 2023 01:15:14 No.7678618 Report

Quoted By: >>7678656

>>7678615
ddim is a meme

Anonymous

Anonymous Tue 28 Nov 2023 01:17:07 No.7678621 Report

Quoted By: >>7678715

>>7678606
DAMN this is almost perfect except for the eyes. Can you correct that somehow? I don't mean generate the entirety again, but just fix the eyes

Anonymous

Anonymous Tue 28 Nov 2023 01:21:19 No.7678629 Report

Quoted By: >>7678636

>>7678617
can i get a catbox for the full res one?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1688707011967761.png, 2MiB, 832x1216

Anonymous Tue 28 Nov 2023 01:21:22 No.7678630 Report

Quoted By: >>7678652

>>7678602
all im losing is cum and time.

Anonymous

Anonymous Tue 28 Nov 2023 01:26:05 No.7678636 Report

Quoted By: >>7678640 >>7678648 >>7679660

>>7678629
for the full res? i mean sure but i think my tinkering would've scrubbed any metadata that it would have had
4chan wont let me do links now for some reason so have the codes before the png for full and base gen
zbvmeq
pdye6p

Anonymous

Anonymous Tue 28 Nov 2023 01:27:09 No.7678640 Report

Quoted By: >>7678648

>>7678636
oh not planning to use it in any way i just really like the end result and i'm autistic about hoarding shit at source quality

Anonymous

Anonymous Tue 28 Nov 2023 01:28:47 No.7678648 Report

Quoted By:

>>7678636
>>7678640
holy shit dude you inpainted the hell out of it, spectacular job

Anonymous

View Same Google ImgOps iqdb SauceNAO 00122-1977222940.png, 464KiB, 1024x1024

Anonymous Tue 28 Nov 2023 01:29:00 No.7678651 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, artist_necro (nekurodayo), (...).jpg, 411KiB, 1856x1280

Anonymous Tue 28 Nov 2023 01:29:25 No.7678652 Report

Quoted By: >>7679402

>>7678602
Really nice gen for local, good work anon.
>>7678630
Same.

Anonymous

View Same Google ImgOps iqdb SauceNAO artist_kinhasu, solo, 1girl, suo (...).png, 2MiB, 1664x1216

Anonymous Tue 28 Nov 2023 01:30:28 No.7678656 Report

Quoted By:

>>7678618
It is a bit of a meme but I like it. Picrel default euler left vs DDIM right

Anonymous

View Same Google ImgOps iqdb SauceNAO ComfyUI_temp_dpzbi_00215_.png, 1MiB, 1024x1024

Anonymous Tue 28 Nov 2023 01:31:59 No.7678661 Report

Quoted By: >>7678683

>>7678613
No but you should use it instead of highres fix.

Anonymous

Anonymous Tue 28 Nov 2023 01:32:58 No.7678665 Report

Quoted By: >>7678668 >>7678672

>>7678602
IF I SUCK A SQUID TIDDY WILL I GET INK OR MILK

Anonymous

Anonymous Tue 28 Nov 2023 01:33:40 No.7678668 Report

Quoted By:

>>7678665
Mink.

Anonymous

Anonymous Tue 28 Nov 2023 01:34:52 No.7678672 Report

Quoted By:

>>7678665
ilk

Anonymous

Anonymous Tue 28 Nov 2023 01:40:54 No.7678679 Report

Quoted By: >>7678680 >>7678819

What's the most underrated tag you know?

Anonymous

Anonymous Tue 28 Nov 2023 01:41:55 No.7678680 Report

Quoted By:

>>7678679
"simple background" in negatives

Anonymous

Anonymous Tue 28 Nov 2023 01:44:03 No.7678683 Report

Quoted By: >>7678708

>>7678661
Oh I see. Is there a link for that?

Anonymous

Anonymous Tue 28 Nov 2023 01:45:05 No.7678686 Report

Quoted By: >>7678693

>>7678602
Do You have a repo of your loras?

Anonymous

Anonymous Tue 28 Nov 2023 01:48:03 No.7678693 Report

Quoted By: >>7678697 >>7678723

>>7678686
i have a mega of ones ive made but im rebaking the marie one rn because of how problematic it's being
the art loras i used are probably all on the gayshit
i kind of want to do another splatoon lora, should i do callie, shiver or frye

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, {{{eromame}}}, solid color (...).png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 01:48:20 No.7678696 Report

Quoted By:

The handling of insertions is really nice.

Anonymous

Anonymous Tue 28 Nov 2023 01:48:47 No.7678697 Report

Quoted By:

>>7678693
you already made the best squid girl, it's over

Anonymous

Anonymous Tue 28 Nov 2023 01:52:44 No.7678705 Report

Quoted By: >>7678707

>>7678391
There’s lots of crazy shit you can’t accomplish without inpainting and DEFINITELY can’t accomplish in 1.5 without hours of inpainting and rerolling. A lot of very undertrained tags for complicated sex positions do not work, period, when prompted for but with some light inpainting will immediately Just Work (tm). Like there’s that particular gangbang position that’s like suspended reverse congress but with two additional dudes flanking the girl on the sides and contributing to the dick count stuffing her holes that is has like less than 5 images on Danbooru depicting it but will still work if you prompt for reverse suspended congress and then inpaint the silhouette of the two extra dudes and their dicks. Or stuff like two girls doing face sitting and cowgirl on a guy at the same time. /d/ stuff like nipple fuck and personality excretion is also a lot easier to inpaint in v3 compared to 1.5 merges.

Or like two threads ago where people were writing out LMAO with rat’s intestines using the x-ray tag using inpainting.

Anonymous

Anonymous Tue 28 Nov 2023 01:54:55 No.7678707 Report

Quoted By: >>7678716 >>7678718

>>7678705
>Or like two threads ago where people were writing out LMAO with rat’s intestines using the x-ray tag using inpainting.
exfuckingcuseme?

Anonymous

Anonymous Tue 28 Nov 2023 01:56:36 No.7678708 Report

Quoted By: >>7678727

>>7678683
Just update your comfy UI and search for PatchModelAddDownscale.

Anonymous

Anonymous Tue 28 Nov 2023 01:58:43 No.7678714 Report

Quoted By:

If I'm generating 1girl 1boy, how do I specify the girl should be winking and not the other character?

Anonymous

View Same Google ImgOps iqdb SauceNAO Alear_AI.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 01:59:10 No.7678715 Report

Quoted By:

>>7678621
Sorry. Not mine.

Anonymous

Anonymous Tue 28 Nov 2023 01:59:11 No.7678716 Report

Quoted By: >>7678722

>>7678707
>>7666142

Anonymous

View Same Google ImgOps iqdb SauceNAO 52B2F8D2-8D6F-417D-BC82-F3B5C7E1 (...).png, 1MiB, 1216x832

Anonymous Tue 28 Nov 2023 01:59:49 No.7678718 Report

Quoted By: >>7678722

>>7678707

Anonymous

Anonymous Tue 28 Nov 2023 02:01:21 No.7678722 Report

Quoted By:

>>7678716
>>7678718
i
i regret asking

Anonymous

Anonymous Tue 28 Nov 2023 02:01:38 No.7678723 Report

Quoted By: >>7678732

>>7678693
Post your mega link plz plz plz also You recently updated the marina one right?

Anonymous

Anonymous Tue 28 Nov 2023 02:03:41 No.7678727 Report

Quoted By: >>7678743

>>7678708
>ComfyUI
Fuck. Of all things...

Anonymous

Anonymous Tue 28 Nov 2023 02:05:32 No.7678732 Report

Quoted By: >>7678750 >>7678767

>>7678723
https://mega.nz/folder/RgkxTRqI#vT0SbjwunzjJA7B9XVr-Lw
yeah i just uploaded the ones ive actually been using
gonna rebake the marina again too, but its pretty good at sex gens, for some reason pinups are iffy plus the clothes arent too accurate, mainly scouring for images with darker skin so she stops genning so goddamn light

Anonymous

Anonymous Tue 28 Nov 2023 02:10:46 No.7678743 Report

Quoted By: >>7678769

>>7678727
I don't use auto11 so I dunno if this is still needed.

https://github.com/wcde/sd-webui-kohya-hiresfix

Anonymous

Anonymous Tue 28 Nov 2023 02:14:14 No.7678750 Report

Quoted By: >>7678758

>>7678732
Anon. Please. Tristana. I'm begging you.

Anonymous

Anonymous Tue 28 Nov 2023 02:16:54 No.7678758 Report

Quoted By: >>7678765

>>7678750
im actually stupid i had a tristana lora i made i forgot to upload, it's up now
it kind of sucks though, way better than the civitai one but maybe set the weight a tad lower if it causes problems

Anonymous

Anonymous Tue 28 Nov 2023 02:22:52 No.7678765 Report

Quoted By:

>>7678758
Bless you.

Anonymous

Anonymous Tue 28 Nov 2023 02:23:59 No.7678767 Report

Quoted By: >>7678794

>>7678732
Anon your character loras are really amazing! I want to ask for details about your way of thinking to improve training, can You elaborate on that? We need more people to train better loras and I want to join the cause, we fellow locals need to ally ourselves these days now, cuz of the rise of NAI

Anonymous

Anonymous Tue 28 Nov 2023 02:25:01 No.7678769 Report

Quoted By:

>>7678743
Thanks I'll give it a try

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_zj2w1w.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 02:29:36 No.7678776 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00164-1075012997.png, 879KiB, 1024x1536

Anonymous Tue 28 Nov 2023 02:31:48 No.7678783 Report

Quoted By: >>7678787 >>7678971

>tfw going through ancient NAI gens from nearly a year ago and cleaning them up, applying styling and upscaling them

Anonymous

Anonymous Tue 28 Nov 2023 02:34:17 No.7678787 Report

Quoted By:

>>7678783
sovl and actually better than aomslop

Anonymous

Anonymous Tue 28 Nov 2023 02:37:56 No.7678794 Report

Quoted By: >>7678846

>>7678767
idunno about all that but its really basic adamw8bit learning/unet 0.0001, text 5e-05, uhhh dim/alpha 128/64, constant 2 3 restarts, 5% warmup, 20 epochs and however many repeats to get what steps i think are appropriate
right now the marie lora has 75 images so im hitting 3000 steps

Anonymous

Anonymous Tue 28 Nov 2023 02:41:30 No.7678797 Report

Quoted By:

>>7678145
based as fuck

Anonymous

View Same Google ImgOps iqdb SauceNAO 1354992118.png, 1MiB, 1216x832

Anonymous Tue 28 Nov 2023 02:41:30 No.7678798 Report

Quoted By: >>7678799 >>7679484 >>7679899

Is this sfw?

Anonymous

Anonymous Tue 28 Nov 2023 02:42:07 No.7678799 Report

Quoted By:

>>7678798
RAPE.

Anonymous

Anonymous Tue 28 Nov 2023 02:45:49 No.7678803 Report

Quoted By: >>7678817 >>7678824

nai doesn't know frieren

:(

Anonymous

View Same Google ImgOps iqdb SauceNAO { {{{character_callie (splatoon) (...).png, 3MiB, 1536x1536

Anonymous Tue 28 Nov 2023 02:47:29 No.7678804 Report

Quoted By: >>7678808

>>7678602
>losing motivation
That's because you chose the inferior sister.

Callie-chads RISE

Anonymous

Anonymous Tue 28 Nov 2023 02:48:09 No.7678806 Report

Quoted By:

>>7678176
based butt cummer

Anonymous

Anonymous Tue 28 Nov 2023 02:49:19 No.7678808 Report

Quoted By: >>7678813

>>7678804
probably true, compiling a dataset for her rn
i like your pic but im too autistic about the characters being on model and too many artists draw splatoon chars nothing like what im going for

Anonymous

Anonymous Tue 28 Nov 2023 02:49:37 No.7678809 Report

Quoted By:

>>7678198
>not being pavlov'd into finding pixel mosaics sexy
ngmi

Anonymous

Anonymous Tue 28 Nov 2023 02:50:45 No.7678810 Report

Quoted By:

>>7678211
post her getting FUCKED by ORCS

Anonymous

View Same Google ImgOps iqdb SauceNAO { {{{character_callie (splatoon) (...).png, 3MiB, 1536x1536

Anonymous Tue 28 Nov 2023 02:53:25 No.7678813 Report

Quoted By: >>7679498

>>7678808
I know how you feel. I had to manually edit her teeth. I can't jerk off to off-model squid teeth, my ancestors would be ashamed.

I'd have had to manually do the cross pupils as well but NAI cleverly sidestepped the issue by making heart pupils instead - which I will allow, but only barely. Close call.

Anonymous

Anonymous Tue 28 Nov 2023 02:54:05 No.7678815 Report

Quoted By:

>>7678488
Euler A because I'm based as fuck

Anonymous

Anonymous Tue 28 Nov 2023 02:55:30 No.7678817 Report

Quoted By: >>7678835 >>7679194

>>7678803
{{frieren}}, {{sousou no frieren}}
green eyes if needed
she had a renaissance towards the end of september when the anime came out, sadly just after NAI got their dataset ready but it definitely knows her

Anonymous

Anonymous Tue 28 Nov 2023 02:55:34 No.7678819 Report

Quoted By:

>>7678679
white background

Anonymous

Anonymous Tue 28 Nov 2023 02:56:34 No.7678822 Report

Quoted By: >>7678984

>>7678585

Getting A1111's inpaint Mask Only feature working is actually what I've been doing off and on today, along getting adetailer to work with it on the txt2img side.

Just managed to get everything functional, and am cleaning things up.

Anonymous

Anonymous Tue 28 Nov 2023 02:57:33 No.7678824 Report

Quoted By:

>>7678803
elf, white hair, twintails,dumb face

Anonymous

View Same Google ImgOps iqdb SauceNAO 2065349346.png, 1MiB, 1216x832

Anonymous Tue 28 Nov 2023 03:00:15 No.7678830 Report

Quoted By: >>7678832 >>7678836 >>7679899

Anonymous

Anonymous Tue 28 Nov 2023 03:02:23 No.7678832 Report

Quoted By: >>7678855

>>7678830
based fizrot enjoyer

Anonymous

View Same Google ImgOps iqdb SauceNAO 1696671053965991.png, 2MiB, 832x1216

Anonymous Tue 28 Nov 2023 03:03:50 No.7678835 Report

Quoted By: >>7678838 >>7679194

>>7678817
Holy shit.

Anonymous

Anonymous Tue 28 Nov 2023 03:04:14 No.7678836 Report

Quoted By: >>7678839 >>7678843 >>7678855 >>7678856

>>7678830
Can you get in trouble for putting a certain word in the prompts?

Anonymous

Anonymous Tue 28 Nov 2023 03:04:56 No.7678838 Report

Quoted By:

>>7678835
see? if the tag exists just {} two or three times
if that's not enough insert the series/copyright and {} as needed

Anonymous

Anonymous Tue 28 Nov 2023 03:05:16 No.7678839 Report

Quoted By: >>7678847

>>7678836
Yes

Anonymous

Anonymous Tue 28 Nov 2023 03:06:06 No.7678843 Report

Quoted By: >>7678847

>>7678836
no, you can't
NAI literally doesn't care and the model can't get you into trouble

Anonymous

Anonymous Tue 28 Nov 2023 03:07:01 No.7678846 Report

Quoted By: >>7678850

>>7678794
Lol I've never got Adamw8bit to work properly, I'll need to test more with it now. Also would I get good results too if I use normal AdamW? I have an AMD gpu and I can't run adamw8bit with it

Anonymous

Anonymous Tue 28 Nov 2023 03:07:04 No.7678847 Report

Quoted By: >>7678851 >>7678852

>>7678839
>>7678843
ARgh who do I believe?

Anonymous

Anonymous Tue 28 Nov 2023 03:09:53 No.7678850 Report

Quoted By:

>>7678846
no clue honestly, i just use google colab

Anonymous

Anonymous Tue 28 Nov 2023 03:10:24 No.7678851 Report

Quoted By:

>>7678847
you quite literally can't generate cee pee, the dataset was assembled from danbooru and not some ukrainian site
that and NAI genuinely don't care about what you do, it's the entire point of the company

Anonymous

Anonymous Tue 28 Nov 2023 03:10:40 No.7678852 Report

Quoted By:

>>7678847
As you might notice, you lose your imagegen history whenever you refresh the page. NAI never keeps anything that isn't encrypted and they never care what you input.

Anonymous

View Same Google ImgOps iqdb SauceNAO 62195039.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 03:11:48 No.7678855 Report

Quoted By: >>7678860 >>7678861 >>7679899

>>7678832
I can't not enjoy the style, it's really good
Excited to start mixing it

>>7678836
What do you mean? They aren't reading prompts if that's what you're implying. The feds WILL use the image metadata against you in court however.

Anonymous

Anonymous Tue 28 Nov 2023 03:12:20 No.7678856 Report

Quoted By: >>7678857

>>7678836
The only way you'll ever get in trouble is pissing off someone powerful.

Anonymous

Anonymous Tue 28 Nov 2023 03:12:59 No.7678857 Report

Quoted By: >>7678863

>>7678856
And at that point it doesn't really matter anyways; they'll just plant stuff.

Anonymous

Anonymous Tue 28 Nov 2023 03:14:00 No.7678860 Report

Quoted By: >>7678863

>>7678855
Wew, dodged a bullet. Thanks for the warning...

Anonymous

Anonymous Tue 28 Nov 2023 03:14:31 No.7678861 Report

Quoted By: >>7678864

>>7678855
Does it work every time for you? All styles do this but fizrot's style in particular tends to melt or just break in very weird ways for me (looks more like a sketch, faces start looking like they were drawn in mspaint, shading just isn't his, etc) and it's very inconsistent.

But when it works... boy does it work.

Anonymous

Anonymous Tue 28 Nov 2023 03:15:33 No.7678863 Report

Quoted By: >>7678867

>>7678860
Like >>7678857 said, if the (((glowniggers))) are after you they'll just plant stuff. They tried doing it earlier this year by releasing a certain model and baiting lolichads into using it.

Anonymous

View Same Google ImgOps iqdb SauceNAO 2153788337.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 03:17:11 No.7678864 Report

Quoted By: >>7678871 >>7678877 >>7679899

>>7678861
It's been working great for me, but I have CFG cranked way up

Try importing my image and using the config

Anonymous

Anonymous Tue 28 Nov 2023 03:18:44 No.7678867 Report

Quoted By:

>>7678863
Reminds me of the dumbass who thought it was a good idea to make mixes with said certain model. I wonder what he's up to now

Anonymous

Anonymous Tue 28 Nov 2023 03:21:57 No.7678871 Report

Quoted By: >>7678873 >>7678877

>>7678864
Will try it, thanks. I just stick to CFG 6 and rescale to 0.5-1. Maybe I should try permanently adding sketch and/or flat shading in my negs and using manual tags to remove the 'very aesthetic'.

Back on local I had a style that I absolutely loved, fizrot+original csrb+takorin all at 0.4, you should be able to see it on aibooru if you look up the tanned cirno tag.

Anonymous

Anonymous Tue 28 Nov 2023 03:22:59 No.7678873 Report

Quoted By: >>7678878

>>7678871
>tanned cirno
Insanely based

Anonymous

Anonymous Tue 28 Nov 2023 03:24:20 No.7678877 Report

Quoted By:

>>7678864
>>7678871
(note that most if not all of the pics genned with that style mix use a different and much worse takorin LoRA)

Anonymous

Anonymous Tue 28 Nov 2023 03:25:23 No.7678878 Report

Quoted By:

>>7678873
Cirno but even better. Thank you ZUN.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 03:34:17 No.7678889 Report

Quoted By: >>7678892 >>7678929

Anonymous

Anonymous Tue 28 Nov 2023 03:37:27 No.7678890 Report

Quoted By: >>7678896

Abandon thread.
Abandon thread.

Anonymous

Anonymous Tue 28 Nov 2023 03:37:39 No.7678892 Report

Quoted By:

>>7678889
cromachina?

Anonymous

Anonymous Tue 28 Nov 2023 03:38:49 No.7678896 Report

Quoted By: >>7678900

>>7678890
?

Anonymous

Anonymous Tue 28 Nov 2023 03:39:43 No.7678900 Report

Quoted By: >>7678922

>>7678896
the artist style

Anonymous

View Same Google ImgOps iqdb SauceNAO 2577711173.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 04:03:08 No.7678922 Report

Quoted By: >>7678928

>>7678900
I dunno, I think it;s pretty good

Anonymous

Anonymous Tue 28 Nov 2023 04:06:32 No.7678928 Report

Quoted By: >>7678938

>>7678922
>also a based Holo enjoyer
By the way what did you reply to? kinda curious

Anonymous

Anonymous Tue 28 Nov 2023 04:06:37 No.7678929 Report

Quoted By: >>7678954

>>7678889
It can do feet up in the air, twitchy motion lines, a reasonably fixable impreg cutout, afterimage/motion blur, better saliva strands, actual penetration (gimped by artist tag), off-centre perspective, directed droplets instead of always straight down

Apart from the fucked up horizontal twitching looking like glitch lines, it's demonstrating a pretty massive gap from 1.5
Also mikusan

Anonymous

View Same Google ImgOps iqdb SauceNAO { {{{character_callie (splatoon) (...).png, 1MiB, 1024x1024

Anonymous Tue 28 Nov 2023 04:09:19 No.7678932 Report

Quoted By:

I'm happy with everything I'm genning except that the eyes and the teeth are not on-model. How fix? I'm already prompting cross-shaped pupils, symbol-shaped pupils.

Anonymous

Anonymous Tue 28 Nov 2023 04:13:07 No.7678938 Report

Quoted By: >>7678950 >>7678958

>>7678928
They seem to think there's an issue with using the artist tag. I don't know anything about it, I just like the style

Anonymous

Anonymous Tue 28 Nov 2023 04:25:38 No.7678950 Report

Quoted By:

>>7678938
For fizrot? Probably just a schizo.

Anonymous

View Same Google ImgOps iqdb SauceNAO {{artist_fizrotart}}, [[artist_h (...).png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 04:29:59 No.7678953 Report

Quoted By: >>7678958

Ignore schizos post more fizrot

Anonymous

Anonymous Tue 28 Nov 2023 04:30:04 No.7678954 Report

Quoted By:

>>7678929
It can do upside down faces

Anonymous

Anonymous Tue 28 Nov 2023 04:35:37 No.7678958 Report

Quoted By: >>7678965 >>7679578

>>7678938
>>7678953
Do I need to be using brackets and parentheses for artist names and certain tags? I'm kinda just putting in words

Anonymous

View Same Google ImgOps iqdb SauceNAO 3662023567.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 04:38:03 No.7678965 Report

Quoted By:

>>7678958
Just b urself
and also try importing the images we post into NAI, since the generation info is stored in the alpha channel. But yeah, a 1 or 2 weights on the artist tag can go a long way.

Anonymous

Anonymous Tue 28 Nov 2023 04:39:49 No.7678967 Report

Quoted By: >>7678968 >>7678970 >>7678981

How to fix broken fingers? So many ruined images that would have otherwise been amazing

Anonymous

Anonymous Tue 28 Nov 2023 04:40:25 No.7678968 Report

Quoted By:

>>7678967
Local? Inpaint.
NAI? Also inpaint.

Anonymous

Anonymous Tue 28 Nov 2023 04:40:50 No.7678970 Report

Quoted By:

>>7678967
Option 1 - Inpaint and hope
Option 2 - Enhance and hope
Option 3 - Use the edit image tool and draw rudimentary fingers and then have it reprocess the image and hope
Option 4 - Re-roll

Anonymous

View Same Google ImgOps iqdb SauceNAO 00249-1284772977.png, 1MiB, 1024x1536

Anonymous Tue 28 Nov 2023 04:41:20 No.7678971 Report

Quoted By:

>wanted to try and avoid having to do any redraw bullshit when going through ancient gens, styling and upscaling
>but ended up putting a decent chunk of work into this one anyway
god dammit

>>7678783
actually, looking at the model hash, the original for that was... AOM3 lmao. Or at least, it matches my aom3 hash.
catbox for the original ancient lowres: https://files.catbox.moe/6bd71q.png
I was doing some real weird shit back then and mostly just trying generalized shit and wasn't actually intentionally going for specific porn.

Anonymous

View Same Google ImgOps iqdb SauceNAO hrtn.png, 276KiB, 521x280

Anonymous Tue 28 Nov 2023 04:51:21 No.7678981 Report

Quoted By: >>7678984 >>7678989 >>7678994 >>7679003 >>7679008

>>7678967
They REALLY need to add inpaint only masked, padding option, and denoising strength to the NAI inpainter. That's all most users need for 99% of inpainting.

For example, the model is more than capable of doing cross-shaped eyes in a close-up (still can't do the funky inkling teeth, but that's sort of on the fact that there's no tag for it and about half or more of artists ignore it), but it's very hard for the NAI inpainting to do get good cross-pupils on inklings on a full sized images. If we could just inpaint only mask in the webpage, this would be a trivial fix that would take 20 seconds. Instead, you have to crop the head, bring it back in, inpaint, paste the inpainted head back onto the original image, img2img to fix inevitable seams from resizing... All such a hassle that it'd be almost as fast to just open up photoshop and draw it yourself if you know how. Very annoying.

Anonymous

Anonymous Tue 28 Nov 2023 04:52:38 No.7678984 Report

Quoted By: >>7678989 >>7679008

>>7678981
>They REALLY need to add inpaint only masked
Based coder anon is working on it.
>>7678822

Anonymous

Anonymous Tue 28 Nov 2023 04:55:41 No.7678989 Report

Quoted By: >>7679000

>>7678981
Yeah, inpainting is massively important for feature refinement or localisation.

>>7678984
It's cool that he is, but this is a paid product, it should have this essential feature.

Anonymous

Anonymous Tue 28 Nov 2023 05:00:19 No.7678992 Report

Quoted By: >>7679004

How do I prompt comic style speech bubbles with precise text?

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{{{ symbol-shaped pupils, cros (...).png, 2MiB, 1024x1024

Anonymous Tue 28 Nov 2023 05:02:06 No.7678994 Report

Quoted By:

>>7678981
Final result BTW.

Remember, literally FUCK every squigger

Anonymous

Anonymous Tue 28 Nov 2023 05:08:31 No.7679000 Report

Quoted By:

>>7678989
>it should have this essential feature
I doubt they care to cater to powerusers. The casual user who cooms to a few txt2imgs once a day is a much more valuable customer.

Anonymous

Anonymous Tue 28 Nov 2023 05:10:33 No.7679003 Report

Quoted By: >>7679020

>>7678981
>All such a hassle that it'd be almost as fast to just open up photoshop and draw it yourself if you know how
literally just takes a $50~ chink pen tablet and a small amount of effort to learn how to do at least a minimal amount of color matching/blending and probably some practice with spot healing brush/clone tool.
It's really not that difficult until you get to the point of having to redraw entire hands or shift anatomy. And once you're used to it it's just plainly significantly better than bothering to try and inpaint anything at all for any reason. Both in terms of not having to try and gacha shit and just plainly having more/better control over how well it meshes in.

Anonymous

Anonymous Tue 28 Nov 2023 05:10:33 No.7679004 Report

Quoted By:

>>7678992
I don't think you can but that's like the easiest thing to manually edit

Anonymous

Anonymous Tue 28 Nov 2023 05:13:46 No.7679008 Report

Quoted By: >>7679025 >>7679038 >>7679148 >>7679511 >>7679591 >>7679835 >>7680108

>>7678585
>>7678981
>>7678984

Just finished adding support for Inpaint only masked to the NAI A1111 extension.

I also re-rigged things so it now works with ADetailer (which requires Inpaint only masked to function with NAI). Might work with some other post-processing extensions, though I haven't tested any (I can't even think of any others off the top of my head)

Anonymous

Anonymous Tue 28 Nov 2023 05:21:38 No.7679020 Report

Quoted By:

>>7679003
>just turn into an artoid
never ever.

Anonymous

Anonymous Tue 28 Nov 2023 05:24:01 No.7679025 Report

Quoted By:

>>7679008
Based, I will be sure to try it later

Anonymous

Anonymous Tue 28 Nov 2023 05:30:32 No.7679037 Report

Quoted By: >>7679051

fyi "artist:omar dogan, {{artist:fizrotart}}" is pretty wild

Anonymous

Anonymous Tue 28 Nov 2023 05:31:02 No.7679038 Report

Quoted By:

>>7679008
Amazing. Literally only thing missing that I regularly use to inpaint is denoise strength.

Anonymous

Anonymous Tue 28 Nov 2023 05:36:08 No.7679050 Report

Quoted By:

Is there a reliable way to just turn a character's hair blonde from black? With just a prompt?

Anonymous

View Same Google ImgOps iqdb SauceNAO 3433320947.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 05:36:39 No.7679051 Report

Quoted By: >>7679061

>>7679037
Just make sure you use CFG 7-10, an DPM++ with SMEA/Dyn

Anonymous

Anonymous Tue 28 Nov 2023 05:36:41 No.7679052 Report

Quoted By: >>7679149

Has anyone gotten takorin to work?
artist:namako (takorin) and pixel art at varying strengths (and maybe pixelated?) kinda start getting the idea

Anonymous

View Same Google ImgOps iqdb SauceNAO 559828670.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 05:44:51 No.7679061 Report

Quoted By:

>>7679051

Anonymous

Anonymous Tue 28 Nov 2023 05:52:44 No.7679070 Report

Quoted By: >>7679077 >>7679608

What's the difference between using [ and {? And how many can I add for each tag? Sorry Im retarded

Anonymous

Anonymous Tue 28 Nov 2023 05:55:29 No.7679077 Report

Quoted By:

>>7679070
[] for deemphasis, {} for emphasis
usually up to 3, maybe 4, any more and it's detrimental
don't use them on every single tag like a retard if you don't need to

Anonymous

Anonymous Tue 28 Nov 2023 06:35:02 No.7679127 Report

Quoted By: >>7679136

API keys are common between imagen and textgen, right? If I somehow manage to scrape a NAI key, I assume I'll get access to both?

Anonymous

Anonymous Tue 28 Nov 2023 06:39:26 No.7679136 Report

Quoted By: >>7679138

>>7679127
Nigger behavior

Anonymous

Anonymous Tue 28 Nov 2023 06:40:05 No.7679138 Report

Quoted By: >>7679140

>>7679136
No regional pricing no buy turk :)

Anonymous

Anonymous Tue 28 Nov 2023 06:40:39 No.7679140 Report

Quoted By: >>7679142

>>7679138
Stop living in a third world country

Anonymous

Anonymous Tue 28 Nov 2023 06:41:27 No.7679142 Report

Quoted By:

>>7679140
>just move to another country bro!
No :)

Anonymous

Anonymous Tue 28 Nov 2023 06:48:08 No.7679148 Report

Quoted By:

>>7679008
Wow, kinda becoming a must-have then. Doesn't it use local GPU for this though? I'm assuming it's cropping the image and inserting it back after?

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{artist_namako (takorin)}}}, { (...).png, 889KiB, 1216x832

Anonymous Tue 28 Nov 2023 06:48:10 No.7679149 Report

Quoted By: >>7679206

>>7679052
best i can do, open to suggestions,

Anonymous

Anonymous Tue 28 Nov 2023 07:11:33 No.7679179 Report

Quoted By: >>7679194

How new is the Nai dataset? Can it draw characters from the latest anime, such as Fern and Frieren?

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{yuuka (blue archive)}}}, {{{b (...).png, 2MiB, 1536x1536

Anonymous Tue 28 Nov 2023 07:14:31 No.7679185 Report

Quoted By:

God damn this is dangerous

Anonymous

Anonymous Tue 28 Nov 2023 07:15:43 No.7679187 Report

Quoted By: >>7679191

>>7678488
DPM++ 2M Karras. Frankly there's so many to choose from I don't know which is "better", or even what the real difference is; all I know is ancestral ones aren't reproducable so I steer away from them

Anonymous

Anonymous Tue 28 Nov 2023 07:18:15 No.7679191 Report

Quoted By: >>7679197

>>7679187
Karras is broken though, stick to the recommended scheduler for 2M

Anonymous

Anonymous Tue 28 Nov 2023 07:19:13 No.7679193 Report

Quoted By: >>7679528 >>7679596

Hol up I'm inserting an image into NAI and trying to generate a very similar looking character in a similar pose. But the result is always nowhere near the original pose or even proper artstyle, even though I have Image2Image at 99 Strength.

Anonymous

Anonymous Tue 28 Nov 2023 07:19:19 No.7679194 Report

Quoted By:

>>7679179
>Frieren
see >>7678817 and >>7678835

Anonymous

Anonymous Tue 28 Nov 2023 07:20:32 No.7679197 Report

Quoted By: >>7679203

>>7679191
>karras is broken
Broken... how? It gens just fine, don't see any big glaring issues. First I've heard of this.

Anonymous

Anonymous Tue 28 Nov 2023 07:22:51 No.7679203 Report

Quoted By:

>>7679197
Leads to broken anatomy and flesh blobs more often than not. YMMV but all the people who tried it ended up switching back to the recommended scheduler, it just works (tm)
That said SDE Karras can lead to some very soft painting-like pictures if that's your thing but again broken anatomy and flesh blobs

Anonymous

View Same Google ImgOps iqdb SauceNAO {{{artist_namako (takorin)}}}, n (...).png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 07:25:09 No.7679206 Report

Quoted By:

>>7679149

Anonymous

Anonymous Tue 28 Nov 2023 07:39:22 No.7679222 Report

Quoted By: >>7679231

why in the goddamn fuck is the sideboob tag forcing a view from the character's left side 95% of the time?

Anonymous

Anonymous Tue 28 Nov 2023 07:44:59 No.7679231 Report

Quoted By: >>7679242

>>7679222
Just flip the image lmao

Anonymous

Anonymous Tue 28 Nov 2023 07:45:24 No.7679232 Report

Quoted By: >>7679241

Where do you guys store your generated files? Is there a good free storage site besides MEGA ?

Anonymous

Anonymous Tue 28 Nov 2023 07:48:27 No.7679241 Report

Quoted By: >>7679247

>>7679232
40$ 1TB external drive

Anonymous

Anonymous Tue 28 Nov 2023 07:48:30 No.7679242 Report

Quoted By:

>>7679231
assuming the character's features are symmetrical sure

Anonymous

Anonymous Tue 28 Nov 2023 07:49:53 No.7679247 Report

Quoted By: >>7679250 >>7679253 >>7679260

>>7679241
I don't want these on my hard drive.....they need to be somewhere secure, reliable but away from my main system

Anonymous

Anonymous Tue 28 Nov 2023 07:51:03 No.7679250 Report

Quoted By:

>>7679247
can you literally not zip them up with a password and hide it in windows' system files?

Anonymous

Anonymous Tue 28 Nov 2023 07:51:58 No.7679253 Report

Quoted By:

>>7679247
I did say external
Just put VeraCrypt on it and store it in the cellar or wherever you want.

Anonymous

Anonymous Tue 28 Nov 2023 07:56:25 No.7679260 Report

Quoted By:

>>7679247
have you considered just not doing something you'd be so fucking afraid of glowies for doing?
And unless you're using actual fucking CSAM with img2img shit random three letter agencies aren't going to give a fuck anyway.
Unless, I guess, if you live in a third world shithole like Canda, UK or Australia and you happen to gen girls that are just a bit too flat or something.

Anonymous

Anonymous Tue 28 Nov 2023 08:13:42 No.7679292 Report

Quoted By: >>7679311

So can we mass upload missing art to Danbooru with the wrong artist tag just so NAI scrapes them up? Sure, they'll be taken down but at least it gets in the database or whatever

Anonymous

Anonymous Tue 28 Nov 2023 08:26:50 No.7679311 Report

Quoted By:

>>7679292
Anon just post a fucking mirror to the dataset you have in /aids/ and tell the NAI devs to pay attention because a shitton of them lurk there. You don't need to go do this roundabout nonsense.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00038-1204632381.png, 2MiB, 1024x1536

Anonymous Tue 28 Nov 2023 08:34:49 No.7679328 Report

Quoted By:

more new old rabbit.

Anonymous

Anonymous Tue 28 Nov 2023 09:35:08 No.7679402 Report

Quoted By: >>7679478

>>7678652
>converting to jpg
WHY

Anonymous

Anonymous Tue 28 Nov 2023 09:42:54 No.7679421 Report

Quoted By: >>7679426

Are there any tags that can add more detail to Nai, similar to the Lora Detail Tweaker?

Anonymous

Anonymous Tue 28 Nov 2023 09:46:17 No.7679426 Report

Quoted By: >>7679445

>>7679421
Just do tiled controlnet on local to the NAI generation.

Anonymous

Anonymous Tue 28 Nov 2023 09:48:04 No.7679429 Report

Quoted By:

>tfw been using detail tweaker at negative values for 50% of my gens
glad I don't have to see this shiny 3d skin ever again man

Anonymous

Anonymous Tue 28 Nov 2023 10:03:27 No.7679445 Report

Quoted By:

>>7679426
What? How

Anonymous

View Same Google ImgOps iqdb SauceNAO ac0ce866-82da-4ead-965d-66f92f44 (...).png, 2MiB, 1856x1280

Anonymous Tue 28 Nov 2023 10:07:23 No.7679451 Report

Quoted By: >>7679500 >>7679592

Do we even need real artists anymore

Anonymous

Anonymous Tue 28 Nov 2023 10:22:11 No.7679478 Report

Quoted By: >>7679487

>>7679402
enjoy your signed .png files

Anonymous

Anonymous Tue 28 Nov 2023 10:23:47 No.7679484 Report

Quoted By:

>>7678798
obviously not fucking retard

Anonymous

Anonymous Tue 28 Nov 2023 10:25:01 No.7679487 Report

Quoted By:

>>7679478
https://github.com/NovelAI/novelai-image-metadata/blob/main/nai_sig.py#L102-L106
It's just a signature over the pixels and comment. Why does it bother you?

Anonymous

Anonymous Tue 28 Nov 2023 10:30:23 No.7679498 Report

Quoted By:

>>7678813
+_+ helps with cross pupils.

Anonymous

Anonymous Tue 28 Nov 2023 10:31:01 No.7679500 Report

Quoted By: >>7681855

>>7679451
yes. there is still a lot the AI can't do yet. gore is partially there but not quite yet for example.

non generic compositions are also a big issue for AI right now

Anonymous

Anonymous Tue 28 Nov 2023 10:35:28 No.7679511 Report

Quoted By:

>>7679008
this is fucking great man, good work

Anonymous

Anonymous Tue 28 Nov 2023 10:39:35 No.7679524 Report

Quoted By:

worst part about prompting nai is that it almost feels like it ignores the emphasis and heavily prioritizes prompt sequence instead. if you want to neg something you pretty much have to disable preset and put the thing you don't want to see before the manually pasted negative

Anonymous

Anonymous Tue 28 Nov 2023 10:41:17 No.7679528 Report

Quoted By:

>>7679193
>99 Strength
Of course it looks nothing like what you put in.

Anonymous

View Same Google ImgOps iqdb SauceNAO the pov hand.png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 10:58:14 No.7679566 Report

Quoted By: >>7679629

Love inpainting

Anonymous

Anonymous Tue 28 Nov 2023 11:04:40 No.7679578 Report

Quoted By: >>7679654

>>7678958
no you dont. CLIP does ai shit on the words, too. you can see that very easily since it will still prompt what you want even if you make a bunch of typos

Anonymous

Anonymous Tue 28 Nov 2023 11:10:07 No.7679588 Report

Quoted By:

Is img2img/inpainting/adetailer paid or are there no limitations as long as it's within 832 x 1216/1216 x 832/1024 x 1024? The
>without any other image used as a base
part is throwing me off since it implies at least img2img and possibly inpainting are paid, but it seems like the only things enforced by the sd-webui's extension's "Force Opus Free Gen Size/Step Limit" option are resolution and size.

Anonymous

Anonymous Tue 28 Nov 2023 11:11:48 No.7679591 Report

Quoted By:

>>7679008
based af

Anonymous

Anonymous Tue 28 Nov 2023 11:12:18 No.7679592 Report

Quoted By:

>>7679451
yes obviously

Anonymous

Anonymous Tue 28 Nov 2023 11:14:02 No.7679596 Report

Quoted By:

>>7679193
blame nai for renaming things like faggots, i assume "strength" is actually denoise, which is the amount of change allowable to the input image. i'm gonna guss 99 is 99% so of course it's not the same

Anonymous

Anonymous Tue 28 Nov 2023 11:19:53 No.7679608 Report

Quoted By:

>>7679070
use the extension and just highlight and press ctrl + up or down

Anonymous

View Same Google ImgOps iqdb SauceNAO LiminalHenya_b.png, 1MiB, 1392x1024

Anonymous Tue 28 Nov 2023 11:29:07 No.7679624 Report

Quoted By: >>7679653

Anonymous

View Same Google ImgOps iqdb SauceNAO LiminalHenya_e.png, 1MiB, 1392x1024

Anonymous Tue 28 Nov 2023 11:29:39 No.7679627 Report

Quoted By: >>7679653

Anonymous

Anonymous Tue 28 Nov 2023 11:30:13 No.7679629 Report

Quoted By:

>>7679566
looks good

Anonymous

View Same Google ImgOps iqdb SauceNAO LiminalHenya_inpaint.png, 1MiB, 1392x1024

Anonymous Tue 28 Nov 2023 11:46:38 No.7679653 Report

Quoted By: >>7680646

>>7679624
>>7679627
can inpaint the hair to make it look nice
also can inpaint the ribs or put ribs in negative to get rid of those
you posted 15min ago so 10min of work to improve image

Anonymous

Anonymous Tue 28 Nov 2023 11:46:47 No.7679654 Report

Quoted By: >>7679655 >>7679667

>>7679578
or give you eyepatches when you ask for "eyepatch bikini"

Anonymous

Anonymous Tue 28 Nov 2023 11:47:34 No.7679655 Report

Quoted By: >>7679656

>>7679654
Good thing they renamed those to square bikini.

Anonymous

Anonymous Tue 28 Nov 2023 11:49:03 No.7679656 Report

Quoted By: >>7679657

>>7679655
Is that a nai only thing? I still see the old tag on danbooru. Well there's a lora for it so w/e

Anonymous

Anonymous Tue 28 Nov 2023 11:50:14 No.7679657 Report

Quoted By:

>>7679656
I meant NAI, yes.

Anonymous

View Same Google ImgOps iqdb SauceNAO masterpiece, best quality, artis (...).png, 1MiB, 1024x1024

Anonymous Tue 28 Nov 2023 11:54:08 No.7679660 Report

Quoted By: >>7679832

>>7678636
>>7678602
My brother, you gotta make a guide. This is amazing.

Anonymous

Anonymous Tue 28 Nov 2023 11:58:18 No.7679667 Report

Quoted By:

>>7679654
c'est la vie

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315410-1602780781-(by (...).png, 1MiB, 1024x1024

Anonymous Tue 28 Nov 2023 12:03:54 No.7679676 Report

Quoted By: >>7679699

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684324769315430-383759483-(by t (...).png, 1MiB, 832x1216

Anonymous Tue 28 Nov 2023 12:17:20 No.7679693 Report

Quoted By: >>7679699

Anonymous

Anonymous Tue 28 Nov 2023 12:18:48 No.7679695 Report

Quoted By: >>7679701

has naiv3 leaked already...?

Anonymous

Anonymous Tue 28 Nov 2023 12:22:30 No.7679699 Report

Quoted By:

>>7679676
>>7679693
better start inpainting those toes

Anonymous

Anonymous Tue 28 Nov 2023 12:23:34 No.7679701 Report

Quoted By: >>7679703

>>7679695
No

Anonymous

Anonymous Tue 28 Nov 2023 12:24:32 No.7679703 Report

Quoted By: >>7679705

>>7679701
y?

Anonymous

View Same Google ImgOps iqdb SauceNAO nsfw, 1girl, 1boy, {robaato}, {{ (...).png, 2MiB, 832x1216

Anonymous Tue 28 Nov 2023 12:25:00 No.7679704 Report

Quoted By: >>7679710 >>7679732

Running out of ideas probably unsubscribing next month

Anonymous

Anonymous Tue 28 Nov 2023 12:25:30 No.7679705 Report

Quoted By:

>>7679703
they took measures to prevent it from happening

Anonymous

View Same Google ImgOps iqdb SauceNAO 22271-2043115333.png, 2MiB, 1280x1600

Anonymous Tue 28 Nov 2023 12:27:10 No.7679708 Report

Quoted By:

>>7678602
dude, same

Anonymous

Anonymous Tue 28 Nov 2023 12:28:40 No.7679710 Report

Quoted By:

>>7679704
I can't stop generating but not sure if I sub next month too, paypigging feels cucked and this shit is not healthy for my addict brain.
But I can't go back to looking at local after this.

Anonymous

Anonymous Tue 28 Nov 2023 12:37:55 No.7679732 Report

Quoted By: >>7679741

>>7679704
>robaato
lmao, just after he ragequit

Anonymous

Anonymous Tue 28 Nov 2023 12:44:48 No.7679741 Report

Quoted By:

>>7679732
dunno who's that, did he quit over ai?

Anonymous

Anonymous Tue 28 Nov 2023 13:00:04 No.7679756 Report

Quoted By:

>>7678560
This post right here is basically the entire synopsis of where local went wrong.

Anonymous

View Same Google ImgOps iqdb SauceNAO image.jpg, 65KiB, 512x768

Anonymous Tue 28 Nov 2023 13:01:53 No.7679759 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_naqve5.png, 2MiB, 1024x1536

Anonymous Tue 28 Nov 2023 13:05:44 No.7679775 Report

Quoted By:

>>7678379
thanks anon, here you go

https://files.catbox.moe/5l4plm.png

Anonymous

View Same Google ImgOps iqdb SauceNAO qqqqqqqqq.jpg, 384KiB, 1216x832

Anonymous Tue 28 Nov 2023 13:05:53 No.7679776 Report

Quoted By: >>7679822 >>7679886

>>7677968

Anonymous

Anonymous Tue 28 Nov 2023 13:39:47 No.7679822 Report

Quoted By: >>7679995

>>7679776
so you decided not to give in to temptation?

Anonymous

Anonymous Tue 28 Nov 2023 13:52:55 No.7679832 Report

Quoted By: >>7681249

>>7679660
thanks man, dunno if i'd know how to make a coherent guide, i kind of just autistically inpaint things i dont like and upscale when i know im not going to get at the little things without a bigger res

Anonymous

Anonymous Tue 28 Nov 2023 13:55:52 No.7679835 Report

Quoted By: >>7679926

>>7679008
Shit, how did you get that to work. I thought you would still have to use some image editing software to splice back in the cut out part at least. Also it would be nice if there was some representation of the area being cut out of the image to send to nai. That way we can be sure it is sending the correct parts with the context it needs. Awesome work.

Anonymous

Anonymous Tue 28 Nov 2023 14:40:11 No.7679878 Report

Quoted By: >>7679882

When Will DALLE 4 be released? So they can release the DALLE 3 model for local

Anonymous

Anonymous Tue 28 Nov 2023 14:42:19 No.7679882 Report

Quoted By: >>7679894 >>7679941

>>7679878
The ClosedAI kikes didn't release DALL-E 2 when DALL-E 3 came out. Neither did NAIggers for NAIv2 when NAIv3 came out. Do you seriously expect cuckporations to cut their profits for the benefits of the people?

Anonymous

Anonymous Tue 28 Nov 2023 14:44:45 No.7679886 Report

Quoted By:

>>7679776
Well shit, now I really want to post more Miku gens. Next bread maybe.

Anonymous

Anonymous Tue 28 Nov 2023 14:50:59 No.7679894 Report

Quoted By: >>7679898

>>7679882
Open source allows models to use their full potential and when NAIv1 leaked it helped massively to improve future models cuz a Lot of people went to adquire experience in finetune, people that were hired for companies.

Anonymous

Anonymous Tue 28 Nov 2023 14:52:53 No.7679898 Report

Quoted By: >>7679907 >>7679911

>>7679894
It's good for AI in general, not for a specific company.

Anonymous

Anonymous Tue 28 Nov 2023 14:53:31 No.7679899 Report

Quoted By: >>7680491

>>7678864
>>7678855
>>7678830
>>7678798
Catbox?

Anonymous

Anonymous Tue 28 Nov 2023 14:58:14 No.7679907 Report

Quoted By: >>7679913

>>7679898
Yes, Indeed. Allows for so much things. Controlnet and it's inpaint feature was created from Open source and it's so good and important that NovelAI integrated the code ok their services. It's thanks to us people of local all over the world that they're not a huge piece of dogshit at what they're doing.

Anonymous

Anonymous Tue 28 Nov 2023 15:00:24 No.7679911 Report

Quoted By: >>7679920

>>7679898
You could make an argument it's better for them too, just in the long run and not as financially good as selling their shit outright. Look at how much shit people came up with, how the methodology for creating stuff matured. They didn't even bother with garbage like hypernetworks because the practice proved them useless, but they're now benefitting or at least considering benefitting from LoRAs and ControlNet. I genuinely think it's worth it for them to always be releasing their previous newest model when a new one comes out for this exact reason.

Anonymous

Anonymous Tue 28 Nov 2023 15:00:59 No.7679913 Report

Quoted By:

>>7679907
Also, You all know that actually NAIv1 is really a piece of garbage compared to nowadays but thanks to many extensions, new checkpoints and inpainting we could make NAIv1 to be usable at least. Even at the level of NAIv3 if You inpaint enough.

Anonymous

Anonymous Tue 28 Nov 2023 15:03:18 No.7679917 Report

Quoted By:

it's over

Anonymous

Anonymous Tue 28 Nov 2023 15:06:00 No.7679920 Report

Quoted By: >>7679923

>>7679911
Yeah. What I meant is it's a competitive advantage against non-AI companies, but not within the AI sphere where everyone gets to benefit from those advancements more or less equally.

Anonymous

Anonymous Tue 28 Nov 2023 15:10:13 No.7679923 Report

Quoted By:

>>7679920
I can cut them some slack for skipping NAIv2, it was a complete fucking nothingburger, but if they aren't gonna release NAIv3 when v4 comes out, all I can conclude is that they're either retarded greedy kikes or just spiteful retards who are STILL mad at the leak and refuse to acknowledge its benefits.

Anonymous

Anonymous Tue 28 Nov 2023 15:11:24 No.7679926 Report

Quoted By: >>7679927 >>7679932

>>7679835
It actually fairly easy, since it's all stuff that A1111 is already doing, I just need to reimplement it in a different context. (The only hard part is python, I've never used it before and keep getting caught up by its atrocious inconsistent syntax.)

The section being sent to NAI is the bounding box containing the mask, expanded by the only masked padding. I may include this fragment in the results, since unlike local, you may actually have paid for that fragment...

Anonymous

Anonymous Tue 28 Nov 2023 15:12:49 No.7679927 Report

Quoted By: >>7679931 >>7679936

>>7679926
For the record, does inpainting cost turkbux?

Anonymous

Anonymous Tue 28 Nov 2023 15:13:48 No.7679931 Report

Quoted By: >>7679933

>>7679927
Depends on your resolution/step count only.

Anonymous

Anonymous Tue 28 Nov 2023 15:14:22 No.7679932 Report

Quoted By:

>>7679926
>I may include this fragment in the results
That would be perfect. Can just do a quick gen to make sure all the context you need is included in the image before setting the batch size and genning a bunch.

Anonymous

Anonymous Tue 28 Nov 2023 15:15:26 No.7679933 Report

Quoted By: >>7679935

>>7679931
Good to know then. This part of the docs threw me off.
>Opus also has a special benefit for Image Generation, allowing you to generate images free of Anlas cost, as long as they're generated one at a time, without any other image used as a base, are at least "Normal Sized", and have a "Steps" value of 28 or less.
>without any other image used as a base

Anonymous

Anonymous Tue 28 Nov 2023 15:16:12 No.7679935 Report

Quoted By:

>>7679933
I think even img2img can be free, as long as the other parameters fit.

Anonymous

Anonymous Tue 28 Nov 2023 15:16:37 No.7679936 Report

Quoted By:

>>7679927
The extension also defaults to enforcing the "free" gen limits, so it shouldn't cost any anime titty funbucks to gen with.

Anonymous

Anonymous Tue 28 Nov 2023 15:21:11 No.7679937 Report

Quoted By:

if the artist prompt makes the pics look fried, try adding "scan, monochrome, greyscale" to negs

Anonymous

Anonymous Tue 28 Nov 2023 15:22:16 No.7679938 Report

Quoted By: >>7679945 >>7679948 >>7679949 >>7679991 >>7679993 >>7680004 >>7680008

>spend a bunch of money on an expensive GPU
>local AI is fucking dead less than a year later
very cool thank you

Anonymous

Anonymous Tue 28 Nov 2023 15:22:46 No.7679941 Report

Quoted By:

>>7679882
This, either we train our own open source model or we submit as pay pigs to the NAItriarchy now and forever, I feel like it it gets released that will outdo their current model and those are lost customers right there.

Anonymous

Anonymous Tue 28 Nov 2023 15:24:37 No.7679945 Report

Quoted By:

>>7679938
I mean you got gaming still right? although if you don't do that yea that tough buddy

Anonymous

Anonymous Tue 28 Nov 2023 15:25:46 No.7679948 Report

Quoted By:

>>7679938
It's just as good as when you bought it, if not better

Anonymous

Anonymous Tue 28 Nov 2023 15:25:52 No.7679949 Report

Quoted By:

>>7679938
you can still generate realistic 1girls standing

Anonymous

Anonymous Tue 28 Nov 2023 15:26:42 No.7679951 Report

Quoted By: >>7679953 >>7679956 >>7679962 >>7679976

>>7678169
>>7678181
IT KEEPS PUTTING BLUE SPANDEX/PANTIES ON HER ASS. I WANT BARE ASS AQUA (only sometimes though but it's still annoying)

Anonymous

Anonymous Tue 28 Nov 2023 15:27:42 No.7679953 Report

Quoted By:

>>7679951
>I WANT BARE ASS AQUA
sigh, don't we all...

Anonymous

Anonymous Tue 28 Nov 2023 15:28:20 No.7679956 Report

Quoted By: >>7679962 >>7680044

>>7679951
such an unreliable model, would never happen with a lora

Anonymous

Anonymous Tue 28 Nov 2023 15:30:16 No.7679960 Report

Quoted By: >>7679963 >>7679969

So....we doing the Danbooru mass upload thing or nah?

Anonymous

Anonymous Tue 28 Nov 2023 15:30:50 No.7679962 Report

Quoted By:

>>7679956
LOL thanks I actually laughed
>>7679951
Copy the prompt? I think I had "no panties" there

Anonymous

Anonymous Tue 28 Nov 2023 15:31:07 No.7679963 Report

Quoted By:

>>7679960
The model was trained once, not continuously. Adding stuff to Danbooru now won't do anything.

Anonymous

Anonymous Tue 28 Nov 2023 15:36:23 No.7679969 Report

Quoted By:

>>7679960
I mean you can if you want v4 to have something specific in like a year, make sure the tags are correct and everything if you do that as well, Im sure you have plenty of time,

Anonymous

Anonymous Tue 28 Nov 2023 15:42:48 No.7679976 Report

Quoted By:

>>7679951
do you have nsfw at front of your prompt? make sure you do

Anonymous

Anonymous Tue 28 Nov 2023 15:51:21 No.7679991 Report

Quoted By:

>>7679938
>not fixing the shit nai3 spits out with local and adding your own loras/style to the image
ishygddt

Anonymous

Anonymous Tue 28 Nov 2023 15:51:50 No.7679993 Report

Quoted By:

>>7679938
OH NO... you can only gen anything you can find or train yourself the horror.

Anonymous

Anonymous Tue 28 Nov 2023 15:55:51 No.7679995 Report

Quoted By: >>7680001

>>7679822
I'd be missing out big time if I didn't try it, just need to set aside the time to get familiar with nai.
Speaking of, what is it the key offers? Is it usage based or time based? I'm planning to extract the max value from it, to that end I've been waiting on the inpaint-masked support to the a1111 addon which has recently been added.

Anonymous

Anonymous Tue 28 Nov 2023 15:58:51 No.7680001 Report

Quoted By: >>7680003

>>7679995
If it's an Opus key, you get 30 days of unlimited gens at normal resolution/28 steps and 10000 anals which can be spent even once the 30 days run out.

Anonymous

Anonymous Tue 28 Nov 2023 16:01:16 No.7680003 Report

Quoted By: >>7680009 >>7680012

>>7680001
damn, I'm being spoiled
how am I supposed to survive 30 days of degenning
anything you want to see made? I've got plenty of ideas for testing things 1.5 is bad at

Anonymous

Anonymous Tue 28 Nov 2023 16:01:35 No.7680004 Report

Quoted By:

>>7679938
I mean if all you were going to do is porn then yeah I guess.

Anonymous

Anonymous Tue 28 Nov 2023 16:04:40 No.7680008 Report

Quoted By:

>>7679938
Sounds like it's time to do animations and cause sour grapes

Anonymous

Anonymous Tue 28 Nov 2023 16:05:55 No.7680009 Report

Quoted By:

>>7680003
>how am I supposed to survive 30 days of degenning
You can only produce so much coom
t. dry after few days

Anonymous

Anonymous Tue 28 Nov 2023 16:07:22 No.7680012 Report

Quoted By: >>7680019

>>7680003
>FFM threesomes (male PoV), especially with bisexual (i.e. kissing/groping/double-dildo) girls
>69
>anal in general
>Yuzusoft style

Anonymous

Anonymous Tue 28 Nov 2023 16:10:46 No.7680019 Report

Quoted By: >>7680028

>>7680012
good taste
never heard of yuzusoft but looks like a VN company with a reasonably modern aesthetic
ffm with girls that enjoy each other too, this is a top tier kink

Anonymous

Anonymous Tue 28 Nov 2023 16:17:44 No.7680028 Report

Quoted By: >>7680045 >>7680050

>>7680019
Also
>tattoos (non-pubic, that shit is easy as fuck)
>fishnets (the bigger the holes the better)
>high heels (particularly toeless)
>undercut hairstyle (on girls)

Anonymous

Anonymous Tue 28 Nov 2023 16:25:04 No.7680044 Report

Quoted By:

>>7679956
this. imagine thinking you'll ever need anything but based64 and loras

Anonymous

Anonymous Tue 28 Nov 2023 16:25:15 No.7680045 Report

Quoted By:

>>7680028
I feel like characters like pokemon's marnie could hit all of these

Anonymous

Anonymous Tue 28 Nov 2023 16:27:00 No.7680048 Report

Quoted By:

Hit the image limit, Old dupe thread is still up.

>>7676460
>>7676460

Anonymous

Anonymous Tue 28 Nov 2023 16:27:40 No.7680050 Report

Quoted By:

>>7680028
>high heels (particularly toeless)
unfortunately V3 seems less capable at making properly shaped high heels than local was, probably due to all the realistic shit mixed in it
unless anyone got some tips

Anonymous

Anonymous Tue 28 Nov 2023 17:04:21 No.7680108 Report

Quoted By: >>7680147

>>7679008
Nice but I've come to report a bug(?)
If I set the sampler and not leave it to "auto sampler" the adetailer images come back as black squares.

Anonymous

Anonymous Tue 28 Nov 2023 17:22:35 No.7680147 Report

Quoted By:

>>7680108
Were you using DDIM? Looks like that isn't supported and always returns an error, even on their own site.

Anonymous

Anonymous Tue 28 Nov 2023 20:03:32 No.7680491 Report

Quoted By: >>7680599

>>7679899
https://files.catbox.moe/ndf5rg.png
https://files.catbox.moe/jd1vcl.png
https://files.catbox.moe/vntiyt.png
https://files.catbox.moe/069t9m.png
https://files.catbox.moe/kltfdk.png
https://files.catbox.moe/ma5eoe.png

Anonymous

Anonymous Tue 28 Nov 2023 20:19:02 No.7680512 Report

Quoted By:

>>7677153
Fuck me, I'd do it for free, or even pay to do it to a horse

Anonymous

Anonymous Tue 28 Nov 2023 20:52:47 No.7680599 Report

Quoted By:

>>7680491
Thanks.

Anonymous

Anonymous Tue 28 Nov 2023 21:10:03 No.7680646 Report

Quoted By:

>>7679653
Thank you Bro. Those are some interesting changes.

Anonymous

Anonymous Wed 29 Nov 2023 02:32:21 No.7681249 Report

Quoted By:

>>7679832
but how? I mean, I inpaint but the eyelashes still come out all fucked for example. How do you achieve such cleanliness?

Anonymous

Anonymous Wed 29 Nov 2023 09:30:10 No.7681855 Report

Quoted By: >>7682498

>>7679500
I exclude gore on my model training and I asume many other also. Propably the reason why gore is not quite well

Anonymous

Anonymous Wed 29 Nov 2023 18:06:37 No.7682498 Report

Quoted By:

>>7681855
you're not training a model, you're a nobody
maybe you mean you're training a LoRA

Anonymous

Anonymous Wed 29 Nov 2023 20:32:16 No.7682732 Report

Quoted By:

Anyone better than me with regional prompter know why I randomly get an extra person or extended heads in my gens? Is it because the image size is too large and it's just trying to fill?
https://files.catbox.moe/5oghop.png

Subject

Name

E-mail

Password

Capcode	All Only User Posts Only Verified Posts Only Moderator Posts Only Manager Posts Only Admin Posts Only Developer Posts Only Founder Posts
Show Posts	All Only With Images Only Without Images Only Spoiler Images Only Non-Spoiler Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

On these archives

On these boards

Your latest searches

/hdg/ - Hentai Diffusion General #387