/h/ - Hentai » Thread #7373269

[949 / 305 / ?]

452KiB, 512x512, 1683884593372071.png

/hdg/ - Hentai Diffusion General #321

Anonymous Sat 13 May 2023 15:19:40 No.7373269 View View Reply Original Report

Quoted By: >>7373483 >>7374102 >>7375050 >>7375530

Lofi Edition

Old: >>7370644

>RESOURCES
Wiki: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki
LoRAs: https://gitgud.io/gayshit/makesomefuckingporn | https://github.com/derrian-distro/LoRA_Easy_Training_Scripts
Tags: https://danbooru.donmai.us/wiki_pages/tag_groups
ControlNet: https://rentry.org/dummycontrolnet
Animation: https://rentry.org/AnimAnon
Wildcards: https://rentry.org/NAIwildcards
4chanX Catbox userscript: https://gist.github.com/catboxanon/ca46eb79ce55e3216aecab49d5c7a3fb
OP Template/Logos: https://pastebin.com/0atr0fEc | https://files.catbox.moe/xk99gy.png | https://files.catbox.moe/35bdt5.gif

Anonymous

View Same Google ImgOps iqdb SauceNAO Trident.jpg, 507KiB, 1536x3072

Anonymous Sat 13 May 2023 15:20:20 No.7373271 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 15:25:13 No.7373279 Report

Quoted By: >>7373283

>>7373228
Please help. I can find a guide anywhere explaining lr/ur rates for different settings.

Anonymous

Anonymous Sat 13 May 2023 15:25:42 No.7373280 Report

Quoted By: >>7373283

>>7373260
underbaked

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765616-3681221510-kill (...).png, 2MiB, 1280x2240

Anonymous Sat 13 May 2023 15:25:58 No.7373282 Report

Quoted By: >>7373315 >>7373624

Anonymous

Anonymous Sat 13 May 2023 15:26:19 No.7373283 Report

Quoted By: >>7373288

>>7373279
If there was a formula, we wouldn't be doing testing.
>>7373280
Oh no, for sure, it was an early epoch.

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765628-2307942498-kill (...).png, 2MiB, 1280x2240

Anonymous Sat 13 May 2023 15:26:35 No.7373284 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 15:31:26 No.7373287 Report

Quoted By:

i got a legible watermark if that's any achievement
https://files.catbox.moe/qjkdkb.png

Anonymous

Anonymous Sat 13 May 2023 15:31:48 No.7373288 Report

Quoted By: >>7373290 >>7373293

>>7373283
alright then what would be responsible for batch 2, alpha 128 because I have no baseline.

Anonymous

Anonymous Sat 13 May 2023 15:32:50 No.7373290 Report

Quoted By: >>7373292 >>7373302

>>7373288
ditching batch2 and using batch1 because its better

Anonymous

Anonymous Sat 13 May 2023 15:34:21 No.7373292 Report

Quoted By: >>7373294

>>7373290
i've heard the exact opposite advice

Anonymous

Anonymous Sat 13 May 2023 15:34:33 No.7373293 Report

Quoted By:

>>7373288
dadapt
but the most common actual lr i see people use is 1e4

Anonymous

Anonymous Sat 13 May 2023 15:35:52 No.7373294 Report

Quoted By: >>7373295 >>7373310 >>7373829

>>7373292
Dunno who told you that, batches are just a speed increase at the cost of quality.

Anonymous

Anonymous Sat 13 May 2023 15:37:38 No.7373295 Report

Quoted By:

>>7373294
The GPUlets were the real master trainers all along...

Anonymous

View Same Google ImgOps iqdb SauceNAO 1683152862749011-2637852896-nsfw (...).png, 1MiB, 768x1152

Anonymous Sat 13 May 2023 15:40:59 No.7373300 Report

Quoted By: >>7374442

Anonymous

Anonymous Sat 13 May 2023 15:41:34 No.7373302 Report

Quoted By: >>7373325

>>7373290
no way, every single thing i've every trained has been better at 2 than 1

Anonymous

View Same Google ImgOps iqdb SauceNAO futa_01_2k.jpg, 2MiB, 2048x3072

Anonymous Sat 13 May 2023 15:44:17 No.7373306 Report

Quoted By: >>7373403

Anonymous

Anonymous Sat 13 May 2023 15:47:46 No.7373310 Report

Quoted By: >>7373325

>>7373294
i assume that someone who doesn't have brain damage told him that

Anonymous

View Same Google ImgOps iqdb SauceNAO justacopy.png, 746KiB, 768x960

Anonymous Sat 13 May 2023 15:49:17 No.7373315 Report

Quoted By: >>7373612

>>7373282
try a corset sometime

Anonymous

Anonymous Sat 13 May 2023 15:51:44 No.7373325 Report

Quoted By: >>7373341 >>7373360

>>7373310
cope

>>7373302
how so?

Anonymous

View Same Google ImgOps iqdb SauceNAO 53b0b694581578155ca6f00d970a6008.jpg, 455KiB, 820x1033

Anonymous Sat 13 May 2023 15:54:48 No.7373334 Report

Quoted By:

How often do you get spooked when proompting?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00024-3214675313-1girl, purple e (...).png, 2MiB, 1280x1792

Anonymous Sat 13 May 2023 15:55:08 No.7373335 Report

Quoted By: >>7373337

Reiq lora will certainly have some good feet.
This took two tries to get.

Anonymous

Anonymous Sat 13 May 2023 15:57:33 No.7373337 Report

Quoted By: >>7373344

>>7373335
Pretty solid artstyle as well, how many images did you use for the lora? I find myself deleting an increasing amount of images every time I try to make an artist lora because I want I want to nudge it towards a specific way the artist draws

Anonymous

Anonymous Sat 13 May 2023 16:00:27 No.7373341 Report

Quoted By: >>7373373 >>7373385

>>7373325
>cope

you're a moron

increased batch sizes have a negative effect on generalization. generalization is a measure of your model's ability to generate images different than your training set. it's not a measure of visual quality or fidelity. it's not even necessarily a bad thing because it's comparable to/identical overfitting. you want that to a certain extent because that's what ensures that your LORA can accurately and consistently recreate your subject.

there is absolutely no reason to run batch size 1 unless you absolutely do not have the VRAM to do higher. you're just losing visual accuracy in exchange for slower training. if you are paranoid about overfitting, or have a smaller dataset, then just use 2 or 3

Anonymous

Anonymous Sat 13 May 2023 16:01:41 No.7373344 Report

Quoted By: >>7373345

>>7373337
>Pretty solid artstyle as well
It is. Maybe not my favorite but it's characteristic.
>how many images did you use for the lora
248 (Prolific! I only realized recently that the guy who does cool Juri feet is also the dude who I jerked off to over a decade ago, heh.). I pruned a good half of the already deduped images. Wanted that specific style feel and also a count below 300.

Anonymous

Anonymous Sat 13 May 2023 16:02:42 No.7373345 Report

Quoted By:

>>7373344
(The other styles I baked were half that, so he really is prolific. Or rather, consistent, since a lot of artists changed their style over the years a LOT.)

Anonymous

Anonymous Sat 13 May 2023 16:05:47 No.7373349 Report

Quoted By:

>>7371951
>>7372473
catbox please

Anonymous

View Same Google ImgOps iqdb SauceNAO 00028-2598108338-1girl, mature f (...).png, 2MiB, 1280x1792

Anonymous Sat 13 May 2023 16:06:25 No.7373350 Report

Quoted By:

>more useful as a foot lora to mix with other styles than the specific foot lora i trained
Smh!
Retromilf + Reiq took 13 rando tries to get good feet.

Anonymous

Anonymous Sat 13 May 2023 16:10:37 No.7373360 Report

Quoted By: >>7373377 >>7373394

>>7373325
it didn't learn the concept i was training it and overfitted on small details of the dataset images

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_yhzjcy.png, 923KiB, 768x1152

Anonymous Sat 13 May 2023 16:11:34 No.7373363 Report

Quoted By: >>7373365 >>7373367 >>7373375 >>7373402

Only starting learning how to use SD today, but I think I'm already getting to be pretty decent at figuring out Lora weights

Anonymous

Anonymous Sat 13 May 2023 16:12:20 No.7373365 Report

Quoted By:

>>7373363
uh sir? your vae?

Anonymous

Anonymous Sat 13 May 2023 16:12:37 No.7373367 Report

Quoted By: >>7373369

>>7373363
vae NOW!!!!!!!!!!!!!!!!!!!!!1
>at figuring out Lora weights
HUH

Anonymous

Anonymous Sat 13 May 2023 16:14:18 No.7373369 Report

Quoted By:

>>7373367
Shit, meant to say additional networks weights*

I'll post the VAEs used in a sec

Anonymous

Anonymous Sat 13 May 2023 16:16:09 No.7373373 Report

Quoted By: >>7373379 >>7373385

>>7373341
>it's not a measure of visual quality or fidelity
>you're just losing visual accuracy
I dont even need to type anything. Though its worth mentioning no one talked about visual quality until you ran yourself into the ground. Go falseflag elsewhere.
Also
>prone to overfitting
>somehow not worse
I dunno if you're actually this dumb or just pretending but this means you can have more steps total without the lora shitting the bed, notable for styles.

Anonymous

Anonymous Sat 13 May 2023 16:17:07 No.7373375 Report

Quoted By: >>7373376 >>7373381

>>7373363
Alright, here's how I did this:

Based64

Front Jack-O Pose:
https://mega.nz/folder/s8wWkZ7C#bu2scWG5kXkarJSN7MKvBg/folder/9s4VnTiC

Sexaroid V7:
https://mega.nz/folder/WjwmjSpA#_Sr_Gpl7WBXwCeSvV0eEtw

Params *should* be embedded in the image if I'm understanding correctly

Anonymous

Anonymous Sat 13 May 2023 16:18:09 No.7373376 Report

Quoted By:

>>7373375
Make sure to have Additional Networks installed, and you may need to rename the Sexaroid V7 safetensor to Sexaroid v7 - see notes for tags - 32dim.safetensors on account of my initial reference for params using that filename.

Anonymous

Anonymous Sat 13 May 2023 16:18:25 No.7373377 Report

Quoted By: >>7373397

>>7373360
>overfitted on small details of the dataset images
tagging issue
>didnt learn the concept
LR issue
dont you have better shit to do than lying on the internet?

Anonymous

Anonymous Sat 13 May 2023 16:19:27 No.7373378 Report

Quoted By:

>another fake war
kys

Anonymous

Anonymous Sat 13 May 2023 16:19:52 No.7373379 Report

Quoted By: >>7373382

>>7373373
if you're going to insist on being stupid then please just stop posting. thanks in advance

Anonymous

Anonymous Sat 13 May 2023 16:20:29 No.7373381 Report

Quoted By: >>7373383

>>7373375
Apply your VAE or I'll stomp your nuts.

Anonymous

Anonymous Sat 13 May 2023 16:21:05 No.7373382 Report

Quoted By:

>>7373379
I accept your concession

Anonymous

Anonymous Sat 13 May 2023 16:21:25 No.7373383 Report

Quoted By: >>7373387

>>7373381
Today I learned what a VAE is, I'm retarded - lemme go do that

Anonymous

Anonymous Sat 13 May 2023 16:24:53 No.7373385 Report

Quoted By: >>7373390 >>7373392

>>7373341
>>7373373
post literally any evidence or I'm ignoring both of you

Anonymous

Anonymous Sat 13 May 2023 16:25:46 No.7373387 Report

Quoted By:

>>7373383
Actually Impressive that you figured out LoRAs before VAEs tho.

Anonymous

Anonymous Sat 13 May 2023 16:28:07 No.7373390 Report

Quoted By: >>7373396

>>7373385
>evidence
This is roleplay and they both haven't trained a LORA since January.

Anonymous

Anonymous Sat 13 May 2023 16:28:29 No.7373392 Report

Quoted By:

>>7373385
If simple logic is beyond you, you can just go do your own testing.

Anonymous

Anonymous Sat 13 May 2023 16:29:02 No.7373394 Report

Quoted By: >>7373400

>>7373360
>it didn't learn the concept i was training

this is the issue with batch size 1. loss is just a measure of how accurately your NN is reproducing the training images on a per pixel level. since changes are only made to the weights when loss goes down this means that it might just never actually be learning what you want it to, and instead be focusing on random one off elements of different pictures. at higher batch sizes the weights are going to be more consistently applied to changes that make a positive impact on each image in the batch, which will make it more likely to be improving the aspect of the image you actually care about.

and even if it does start to figure something it struggles with out at batch size 1 it's more likely to run into catastrophic interference.

just never do it. slight overfitting is always gonna be better than significant underfitting if you're trying to generate anime waifus

Anonymous

Anonymous Sat 13 May 2023 16:30:23 No.7373396 Report

Quoted By: >>7373399

>>7373390
>since January
Since when were you under the impression that I trained any in the first place?

Anonymous

Anonymous Sat 13 May 2023 16:31:02 No.7373397 Report

Quoted By:

>>7373377
lole

Anonymous

Anonymous Sat 13 May 2023 16:32:52 No.7373399 Report

Quoted By: >>7373401

>>7373396
>singular lora posted in the last five threads
you know
this is not even shitposting at this point

Anonymous

Anonymous Sat 13 May 2023 16:33:44 No.7373400 Report

Quoted By: >>7373414

>>7373394
>at higher batch sizes the weights are going to be more consistently applied to changes that make a positive impact on each image in the batch
lol you're actually just full blown larping at this point
The issues you described aint real, I've never had them happen despite training since january.

Anonymous

Anonymous Sat 13 May 2023 16:35:44 No.7373401 Report

Quoted By: >>7373415

>>7373399
>ignoring all loras posted on civit just because someone didnt link a mega
there was 10+ mentioned itt the past week

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_0bu1sz.png, 792KiB, 768x1152

Anonymous Sat 13 May 2023 16:36:02 No.7373402 Report

Quoted By:

>>7373363
Take two, VAE applied - why did I decide to fuck around with this while running on fumes.

Anonymous

Anonymous Sat 13 May 2023 16:36:07 No.7373403 Report

Quoted By: >>7373409 >>7373418 >>7373421

>>7373306
jannies actually sleeping

Anonymous

Anonymous Sat 13 May 2023 16:37:16 No.7373405 Report

Quoted By:

OH FUCK, the inpaint with latent upscale only works if you have sent the image from txt2img to inpaint? So it needs to have the image in either txt/img2img to work? When I don't have that, it says missing image.

Anonymous

Anonymous Sat 13 May 2023 16:40:02 No.7373409 Report

Quoted By:

>>7373403
they're na

Anonymous

View Same Google ImgOps iqdb SauceNAO 00216-3748100941.png, 967KiB, 768x1152

Anonymous Sat 13 May 2023 16:41:01 No.7373411 Report

Quoted By:

Damn, applying that VAE really did make a world of difference, thanks for letting me know I was acting like a lobotomite

Anonymous

View Same Google ImgOps iqdb SauceNAO 01049-3680182783.png, 931KiB, 602x902

Anonymous Sat 13 May 2023 16:42:03 No.7373412 Report

Quoted By: >>7374643

Anonymous

Anonymous Sat 13 May 2023 16:43:08 No.7373414 Report

Quoted By: >>7373443

>>7373400
>lol you're actually just full blown larping at this point

if it isn't calculating the gradients for each image in a mini batch for every step then how exactly does training work

Anonymous

Anonymous Sat 13 May 2023 16:43:10 No.7373415 Report

Quoted By: >>7373420

>>7373401
>civit
>ignore
yeah

Anonymous

View Same Google ImgOps iqdb SauceNAO 00011-1289187061.png, 2MiB, 1024x1536

Anonymous Sat 13 May 2023 16:44:09 No.7373417 Report

Quoted By: >>7373426 >>7373455

reminder that anyone who makes broad claims and refuses to substantiate them is full of shit. Doubly so if they expect you to substantiate their claims for them

Anonymous

Anonymous Sat 13 May 2023 16:44:20 No.7373418 Report

Quoted By: >>7373426 >>7373430

>>7373403
is that image against the rules or what?

Anonymous

Anonymous Sat 13 May 2023 16:44:43 No.7373420 Report

Quoted By: >>7373424

>>7373415
Get used to it. Theres no point posting megas outside of lolis and even then you can get nuked

Anonymous

Anonymous Sat 13 May 2023 16:45:16 No.7373421 Report

Quoted By:

>>7373403
cringe faggots awake though

Anonymous

Anonymous Sat 13 May 2023 16:45:23 No.7373422 Report

Quoted By: >>7373425

At this point I only seriously train a LoRA if I find a new character or style I like, which is not often because I barely consoom media anymore; everything old I already have good enough LoRAs for. The rest of the time I just cook old datasets on newly added or generally retarded settings to find out what they do. i.e. currently trying out automated resize -> --dim_from_weights training

Anonymous

Anonymous Sat 13 May 2023 16:45:43 No.7373424 Report

Quoted By:

>>7373420
I'm not posting my shit on that garbage site

Anonymous

Anonymous Sat 13 May 2023 16:46:15 No.7373425 Report

Quoted By:

>>7373422
I'm just baking for the sake of it nowadays.
I don't use styles, but it's fun to see how they work.
And same, I have my roster of characters that's unlikely to change soon.

Anonymous

Anonymous Sat 13 May 2023 16:46:16 No.7373426 Report

Quoted By: >>7373432 >>7373435 >>7373460

>>7373418
futa is /d/, weirdly enough

>>7373417
batches hurting quality isn't something new and people have posted grids of it happening way back in february, there were even papers linked on this subject newfag

Anonymous

Anonymous Sat 13 May 2023 16:47:06 No.7373430 Report

Quoted By:

>>7373418
even the legally blind can have filenames read to them

Anonymous

Anonymous Sat 13 May 2023 16:47:51 No.7373432 Report

Quoted By:

>>7373426
>there were even papers linked on this subject newfag

you should have tried actually reading them when i posted them

Anonymous

Anonymous Sat 13 May 2023 16:48:40 No.7373434 Report

Quoted By:

>they're still larping
>they still aren't posting images or loras or anything else
KEK!

Anonymous

Anonymous Sat 13 May 2023 16:48:40 No.7373435 Report

Quoted By: >>7373448

>>7373426
damn should be pretty easy for you to post literally anything then

Anonymous

Anonymous Sat 13 May 2023 16:52:13 No.7373443 Report

Quoted By:

>>7373414
Never said anything about this, are you ok? Anon was just making out the ass claims on everything else
>only adjusts weights when loss goes down
a single issue would kill all loras then because it would never adjust as the loss continued to spike up
>random one off element
doesnt happen because tagging exists
>weights more consistently applied, more likely to be improving
lol, lmao, would like to know how anon got inside the black box and checked things.

Anonymous

Anonymous Sat 13 May 2023 16:53:52 No.7373448 Report

Quoted By: >>7373455

>>7373435
as easy as it should be for you to dive into archive or test it yourself. if you want me to spoon feed you then pay me for the trouble newfaggot, I ain't your victorian maid.

Anonymous

Anonymous Sat 13 May 2023 16:56:40 No.7373454 Report

Quoted By:

This is pretty pathetic for a general as dead as this desu

Anonymous

Anonymous Sat 13 May 2023 16:57:21 No.7373455 Report

Quoted By: >>7373461

>>7373448
>>7373417

Anonymous

Anonymous Sat 13 May 2023 16:58:53 No.7373460 Report

Quoted By: >>7373470

>>7373426
>there were even papers linked on this subject newfag

yeah here is the paper that you're probably talking about since it's one that I linked a few months ago

https://www.semanticscholar.org/reader/03cf148638e007ddb42ac49f91225712b6c66a08

>The presented results confirm that using small batch sizes achieves the best training stability and
generalization performance, for a given computational cost, across a wide range of experiments. In
all cases the best results have been obtained with batch sizes m = 32 or smaller, often as small as
m = 2 or m = 4.

it does not say to use batch size 1.

Anonymous

Anonymous Sat 13 May 2023 16:58:56 No.7373461 Report

Quoted By:

>>7373455
Cope all you want, still not spoonfeeding you newfag. Google it.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1683152862749015-1236994578-full (...).png, 1MiB, 768x1152

Anonymous Sat 13 May 2023 17:02:13 No.7373468 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 17:04:34 No.7373470 Report

Quoted By: >>7374646

>>7373460
Because they never tested on 1 you imbecile. 2 was the smallest number they went with, you can even see it on the fucking graphs. Maybe you should read the papers you link.

Anonymous

Anonymous Sat 13 May 2023 17:06:49 No.7373477 Report

Quoted By:

janny...

Anonymous

View Same Google ImgOps iqdb SauceNAO 00828-mistoonAnime_v10_1261780865.png, 924KiB, 768x768

Anonymous Sat 13 May 2023 17:08:57 No.7373483 Report

Quoted By: >>7373490

>>7373269

Hey thats MINE!!!

Thanks for using it btw.

Anonymous

View Same Google ImgOps iqdb SauceNAO 01440-3857334034.png, 1MiB, 1024x1344

Anonymous Sat 13 May 2023 17:09:25 No.7373486 Report

Quoted By: >>7373491 >>7373497

controlnet is the future with tile upscale
so easy to do, no more rng gens
and you can fuzzy it to get art that is inspired by other art (0.7 denoise on controlnet)
controlnet used to be bad but it is actually quite good now with preprocessor: lineart_anime
sometimes touch ups are required on the preprocessed image with inpainting, this is only if u want to maximize what u are after.
i suggest if you find a cool scene in a doujin, cut out the panel and preprocess it, clean up the preprocessed image then create your gens.

controlnet
>https://github.com/Mikubill/sd-webui-controlnet
tile upscaling (read the issues page to learn how to use it)
>https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111

Anonymous

Anonymous Sat 13 May 2023 17:10:51 No.7373490 Report

Quoted By:

>>7373483
It's kino.
Better than anything that's been posted in a while by these incessant bitches.

Anonymous

Anonymous Sat 13 May 2023 17:12:18 No.7373491 Report

Quoted By: >>7373495 >>7373511 >>7373528 >>7373558

>>7373486
ControlNet was actually updated with a new reference-only preprocessor a few hours ago
I'm going to check it out and see if it's any good
https://github.com/Mikubill/sd-webui-controlnet/discussions/1236

Anonymous

View Same Google ImgOps iqdb SauceNAO 00194-4186779332.png, 2MiB, 1536x1024

Anonymous Sat 13 May 2023 17:14:41 No.7373494 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00762-891154882-tatsumaki_1girl_ (...).png, 1MiB, 1024x1536

Anonymous Sat 13 May 2023 17:15:27 No.7373495 Report

Quoted By: >>7373528

>>7373491
ControlNet author keeping this general alive single handedly

Anonymous

Anonymous Sat 13 May 2023 17:17:16 No.7373497 Report

Quoted By: >>7373499 >>7373528

>>7373486
I just found out about the tile processor. I'm running into the usual problem of the image becoming chaos, cloning stuff everywhere. Am I going too high res in one go?

Anonymous

Anonymous Sat 13 May 2023 17:19:54 No.7373499 Report

Quoted By: >>7373507 >>7373527

>>7373497
Probably has to do with tile size etc, since it upscales the image in parts, so if the parts are too small or there's not enough padding it will do this

Anonymous

View Same Google ImgOps iqdb SauceNAO 00445-mistoonSapphire_v10_341803 (...).png, 512KiB, 512x768

Anonymous Sat 13 May 2023 17:25:18 No.7373505 Report

Quoted By: >>7373560

Let me retro post some more. Speaking of, anyone like the boomer fat daddy vibes?

Anonymous

Anonymous Sat 13 May 2023 17:25:42 No.7373507 Report

Quoted By: >>7373519

>>7373499
Can't see any settings to control the amount of tiles though.

Anonymous

View Same Google ImgOps iqdb SauceNAO tmp6vrdp1fu.png, 301KiB, 512x512

Anonymous Sat 13 May 2023 17:25:53 No.7373508 Report

Quoted By: >>7373513 >>7373514 >>7373516 >>7373517 >>7373521 >>7373523 >>7373531 >>7373532 >>7373536 >>7373543 >>7373877 >>7373899

Man you guys get actual art and I'm still just generating bullshit. What's the secret? I followed the speedrun setup guide but this is probably the best image out of about an hour of prompting

Anonymous

Anonymous Sat 13 May 2023 17:27:35 No.7373511 Report

Quoted By: >>7373515 >>7373535 >>7374063 >>7374066

>>7373491
If you combine this with the shuffle processor it's almost a one-shot style transfer, pretty sick

Anonymous

Anonymous Sat 13 May 2023 17:28:21 No.7373513 Report

Quoted By:

>>7373508
soul

Anonymous

Anonymous Sat 13 May 2023 17:28:32 No.7373514 Report

Quoted By: >>7373520

>>7373508
>speedrun guide
don't know of any such things

Anonymous

Anonymous Sat 13 May 2023 17:28:37 No.7373515 Report

Quoted By: >>7373535

>>7373511
*shuffle controlnet
Both need the strength lowered a bit too if you want to combine them

Anonymous

Anonymous Sat 13 May 2023 17:29:28 No.7373516 Report

Quoted By: >>7373520

>>7373508
What model, sampler, steps

Anonymous

Anonymous Sat 13 May 2023 17:30:05 No.7373517 Report

Quoted By: >>7373520

>>7373508
The original novelai model is pretty old at this point and most people have moved on to better models
Loras can also be used to essentially import characters and artstyles that the models aren't already familiar with
>512x512
Enable hires fix

Anonymous

View Same Google ImgOps iqdb SauceNAO file.png, 3KiB, 547x31

Anonymous Sat 13 May 2023 17:31:27 No.7373519 Report

Quoted By: >>7373525

>>7373507
its the tile sizes

Anonymous

Anonymous Sat 13 May 2023 17:33:12 No.7373520 Report

Quoted By: >>7373526 >>7373529 >>7373530 >>7373534

>>7373514
The one in the OP of this thread
>>7370644

>>7373516
animefull-latest
and the vae included with the novelAI I guess
35 steps
Whatever the default sampler is

>>7373517
What new models are there? is it all in the OP? I'm still new at this

Anonymous

Anonymous Sat 13 May 2023 17:33:17 No.7373521 Report

Quoted By:

>>7373508
dpm++2m karras or SDE sampler, 25-30 steps for 2m, less for SDE, highres fix at 0.5 denoise on ESRGAN or animesharp. Loras and models that aint NAI

Anonymous

View Same Google ImgOps iqdb SauceNAO 00253-icomix_V04_1469447561.png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 17:33:46 No.7373523 Report

Quoted By:

>>7373508

My 2060 could only take so much. I switched to colab. Nocrypt's one. It good desu.

Anonymous

Anonymous Sat 13 May 2023 17:35:13 No.7373525 Report

Quoted By: >>7373534

>>7373519
I might be retarded, this is not the tile_resample in controlnet?

Anonymous

Anonymous Sat 13 May 2023 17:35:34 No.7373526 Report

Quoted By:

>>7373520
Default is Euler A
Don't use that
Use one of the karras ones

Anonymous

Anonymous Sat 13 May 2023 17:36:06 No.7373527 Report

Quoted By: >>7373534

>>7373499
what tile size do you recommend

Anonymous

View Same Google ImgOps iqdb SauceNAO 00448-6117337332.png, 1MiB, 1024x944

Anonymous Sat 13 May 2023 17:36:38 No.7373528 Report

Quoted By:

>>7373491
good luck anon, i recommend trying it with loras you will be surprised
>>7373495
kek true
>>7373497
normally you want denoise low, if denoise goes high you lose the original image. start at low denoise and slowly increase it, and try it with a small image so you dont have to wait ages between testing

Anonymous

Anonymous Sat 13 May 2023 17:36:42 No.7373529 Report

Quoted By:

>>7373520
Lots of people in these generals use various AbyssOrangeMix (abbreviated as AOM) or BasedMix which itself also is a model mix based on AOM
Others use their own models since you can merge existing models and play around with different weight scalings yourself
You can find the models alongside various other models on Huggingface

Anonymous

Anonymous Sat 13 May 2023 17:36:56 No.7373530 Report

Quoted By:

>>7373520
Do DPM++ SDE Karras on 20-25 steps

Anonymous

Anonymous Sat 13 May 2023 17:37:23 No.7373531 Report

Quoted By: >>7373533

>>7373508
>actual art

Anonymous

Anonymous Sat 13 May 2023 17:38:24 No.7373532 Report

Quoted By:

>>7373508
the secret is a combination of secret sauce mixing, loras, and extensions. you can download all these without building them yourself.

Anonymous

Anonymous Sat 13 May 2023 17:38:33 No.7373533 Report

Quoted By: >>7373537

>>7373531
cope

Anonymous

View Same Google ImgOps iqdb SauceNAO file.png, 3KiB, 252x72

Anonymous Sat 13 May 2023 17:38:37 No.7373534 Report

Quoted By: >>7373540 >>7373548

>>7373520
https://civitai.com/ for models/loras, if not there see https://huggingface.co/
AbyssOrangeMix2 (AOM2), based64, 7thheaven are ok models to start with. Download only safetensors files, .ckpt is risky.
>VAE
see pic related for decent ones, theres prob more but I dont follow them much - base NAI one is desatured and buggy.

>>7373525
>tile upscaling (read the issues page to learn how to use it)
>https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111
Tile upscaling is not tile preprocesssor I imagine, not sure what controlnet's tile stuff is.

>>7373527
dunno I dont use it, I dont gen large images usually and its worse than highresfix because of the tile thing. read the github and play with it till it works.

Anonymous

Anonymous Sat 13 May 2023 17:40:34 No.7373535 Report

Quoted By: >>7374695

>>7373511
>>7373515
bro any screencap for settings?, been a while since I've used controlnet and I only used depth and cunny or whatever it's called

is stuff like open pose still useful

Anonymous

Anonymous Sat 13 May 2023 17:40:37 No.7373536 Report

Quoted By:

>>7373508
Well I have around 10k generated Images so I would consider practice an important part of the process

Anonymous

Anonymous Sat 13 May 2023 17:41:36 No.7373537 Report

Quoted By:

>>7373533
projecting

Anonymous

View Same Google ImgOps iqdb SauceNAO xczzxc.jpg, 1MiB, 1920x2880

Anonymous Sat 13 May 2023 17:43:53 No.7373539 Report

Quoted By: >>7373541 >>7373660 >>7373764

wow multidiffusion is working great

Anonymous

Anonymous Sat 13 May 2023 17:45:50 No.7373540 Report

Quoted By:

>>7373534
uh ok

Anonymous

Anonymous Sat 13 May 2023 17:48:39 No.7373541 Report

Quoted By: >>7373544 >>7373557

>>7373539
kek
the regulars here hide their secrets, multidiffusion upscaler is one of them. you can think of high res fix as v1, standard img2img upscale as v2, and multidiffusion upscaler as v3

Anonymous

Anonymous Sat 13 May 2023 17:49:39 No.7373543 Report

Quoted By: >>7373564 >>7373573

>>7373508
>actual art
share which 3 in this thread are closest to actual art, each time I browse I see endless issue through which you can recognize its ai generated stuff on top of overdetailed aspects which ruin the piece usually

Anonymous

Anonymous Sat 13 May 2023 17:51:53 No.7373544 Report

Quoted By: >>7373551

>>7373541
what secrets

Anonymous

View Same Google ImgOps iqdb SauceNAO 01103-4181637465-outstretched_ar (...).png, 1MiB, 896x1184

Anonymous Sat 13 May 2023 17:54:12 No.7373547 Report

Quoted By: >>7373577

Anonymous

Anonymous Sat 13 May 2023 17:54:21 No.7373548 Report

Quoted By:

>>7373534
Ah, I'm merely retarded then, thanks anon, this helps a ton!

Anonymous

View Same Google ImgOps iqdb SauceNAO 00109-2992806911-1girl_1boy_side (...).png, 2MiB, 1536x1024

Anonymous Sat 13 May 2023 17:54:52 No.7373549 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00820-4130819614.png, 3MiB, 1536x2184

Anonymous Sat 13 May 2023 17:55:20 No.7373551 Report

Quoted By: >>7373553 >>7373567 >>7373571

>>7373544
not sharing secrets. pretending catbox prompts will get you their images with no extra work

Anonymous

Anonymous Sat 13 May 2023 17:58:50 No.7373553 Report

Quoted By:

>>7373551
I don't believe you.

Anonymous

Anonymous Sat 13 May 2023 17:58:51 No.7373554 Report

Quoted By: >>7373729

>>7373003
catbox?

Anonymous

Anonymous Sat 13 May 2023 18:00:32 No.7373557 Report

Quoted By: >>7373559

>>7373541
schizo projection

Anonymous

View Same Google ImgOps iqdb SauceNAO 21276-1681793134-1girl, shirt li (...).png, 731KiB, 768x1152

Anonymous Sat 13 May 2023 18:02:24 No.7373558 Report

Quoted By: >>7373585 >>7373642

>>7373491
How well does it handle 2D drawings as reference images?
I tried using one as a reference image and it gave me something completely different that doesn't match the pose in the reference pic at all. I tried turning controlnet off and remaking the pic with the same pic and it came out different once again, so it's evidently doing something. It just doesn't seem to be as strong as the example pictures in the github repo makes it out to be

Anonymous

Anonymous Sat 13 May 2023 18:02:30 No.7373559 Report

Quoted By: >>7373561

>>7373557
im not hiding what im doing so where is the projection
schizo is the zoomer newfag

Anonymous

Anonymous Sat 13 May 2023 18:04:06 No.7373560 Report

Quoted By:

>>7373505
who's the girl char?

Anonymous

Anonymous Sat 13 May 2023 18:04:15 No.7373561 Report

Quoted By:

>>7373559
Well no one is hiding stuff then and I'll grant you that there's no projection - you're just a schizo.

Anonymous

Anonymous Sat 13 May 2023 18:07:01 No.7373564 Report

Quoted By:

>>7373543
i think youve just spent to much time looking at decent AI art is all. go back and look at some hand drawn shit for a while from random artists, then compare notes. i'm honestly surprised by the hand gen results sometimes compared to the source material these models were trained on....

Anonymous

Anonymous Sat 13 May 2023 18:08:09 No.7373567 Report

Quoted By:

>>7373551
I mean they usually say when they use img2img or controlnet

Anonymous

Anonymous Sat 13 May 2023 18:09:52 No.7373571 Report

Quoted By:

>>7373551
use the secrets to fix her goblin chin pls

Anonymous

Anonymous Sat 13 May 2023 18:10:10 No.7373573 Report

Quoted By:

>>7373543
it's some schizo projector

Anonymous

View Same Google ImgOps iqdb SauceNAO 00023-1869707957.png, 2MiB, 1280x1280

Anonymous Sat 13 May 2023 18:11:27 No.7373576 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 18:11:57 No.7373577 Report

Quoted By: >>7373579 >>7373580

>>7373547 (me)
btw, any tips to get hair/necklaces hanging down instead of ignoring gravity?

Anonymous

Anonymous Sat 13 May 2023 18:13:38 No.7373578 Report

Quoted By: >>7373580

uegh so tired
don't even want to test epochs

Anonymous

Anonymous Sat 13 May 2023 18:14:50 No.7373579 Report

Quoted By: >>7373591

>>7373577
that is hanging down, if it weren't she'd fall. the camera isn't oriented normal to the earth's surface

Anonymous

Anonymous Sat 13 May 2023 18:14:58 No.7373580 Report

Quoted By: >>7373591 >>7373595

>>7373577
>detailed
? not sure what ur lookin for. Shading? idk

>>7373578
im stuck halfway through re-doing tags. Maybe I should raise from 0.15 to have less work

Anonymous

View Same Google ImgOps iqdb SauceNAO 21292-2762827312-1girl, shirt li (...).png, 633KiB, 768x1152

Anonymous Sat 13 May 2023 18:21:24 No.7373585 Report

Quoted By: >>7373589 >>7373642

>>7373558
Playing around with it some more and it seems decent enough at transferring the style of the reference picture provided you prompt what's already in the picture; the more you deviate from describing the picture the less it resembles it.
I know that sounds really obvious when it's worded like this but I was just hoping it could be used as an improved ControlNet if you wanted to just copy a single pose from an existing picture or something similar. It's a really good tool none the less, you anons should try using it with loras as well

Anonymous

View Same Google ImgOps iqdb SauceNAO 00044-4233060729.png, 1MiB, 1152x1280

Anonymous Sat 13 May 2023 18:23:53 No.7373587 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 18:25:17 No.7373589 Report

Quoted By: >>7373593

>>7373585
The art style looks perfect catbox?

Anonymous

Anonymous Sat 13 May 2023 18:26:24 No.7373591 Report

Quoted By: >>7373601 >>7373633

>>7373579
She isn't standing anon, the view is supposed to be from below and she's supposed to be leaning over the camera. The floor isn't visible there at all, it's supposed to be a view of the sky and an arena wall.
I can see how it didn't come across well on this seed though.
Anyway, basically I want the hair and necklace to fall down toward the camera and hang but idk how. Hair I think works when you prompt it longer but the necklace seems like a lost cause.
>>7373580
wut?

Anonymous

Anonymous Sat 13 May 2023 18:26:46 No.7373593 Report

Quoted By:

>>7373589
https://files.catbox.moe/zvu42d.png
The source pic I used is https://www.pixiv.net/en/artworks/87480539

Anonymous

Anonymous Sat 13 May 2023 18:27:05 No.7373595 Report

Quoted By: >>7373601

>>7373580
>im stuck halfway through re-doing tags
I gaslit myself into thinking EU4 is fun again, so I don't want to get away. I should've put in the other loras to bake in the background now that I think about it though.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00049-1664014582.png, 2MiB, 1152x1280

Anonymous Sat 13 May 2023 18:28:01 No.7373597 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 18:31:15 No.7373601 Report

Quoted By: >>7373603 >>7373607

>>7373591
>wut
what? You want details, shading to signify its actually not glued to her body. Try inpainting the area with stuff like "hanging from neck" or whatever the tags are.

>>7373595
are you literally me? Been having fun with Eu4 for a while now again, absolutely great patches recently.
>bake in the BG
EU is reliant on GPU from my experience - last time I tried that it was crawling.

Anonymous

Anonymous Sat 13 May 2023 18:33:33 No.7373603 Report

Quoted By: >>7373613

>>7373601
>Been having fun with Eu4 for a while now again, absolutely great patches recently.
Yeah, it really has been.
I still have a habit of ending around 1600 but in general it seems to be more fun.
>EU is reliant on GPU from my experience - last time I tried that it was crawling.
I did it before. Low batches and plenty of headroom makes it smooth.

Anonymous

Anonymous Sat 13 May 2023 18:35:52 No.7373605 Report

Quoted By: >>7373609 >>7373613 >>7373620

>See Automatic1111 updated to a new version
>Figure it probably fixed most of the things that were fucked with the previous versions
>git pull
>Loras now only apply the final version if you have multiple versions of the same lora (IE different epochs)
>Can no longer quick select all of them through the extra networks pop-up
Literally and I do mean LITERALLY zero reason for this change to exist

Anonymous

Anonymous Sat 13 May 2023 18:36:49 No.7373606 Report

Quoted By: >>7373608

>OutOfMemoryError: CUDA out of memory. Tried to allocate 36.00 GiB (GPU 0; 24.00 GiB total capacity; 22.21 GiB already allocated;

AUGH

Anonymous

Anonymous Sat 13 May 2023 18:38:24 No.7373607 Report

Quoted By:

>>7373601
I will try it, though there is a big skill issue when it comes to me and inpainting.

Anonymous

Anonymous Sat 13 May 2023 18:38:28 No.7373608 Report

Quoted By:

>>7373606
ACK

Anonymous

Anonymous Sat 13 May 2023 18:38:49 No.7373609 Report

Quoted By: >>7373611

>>7373605
pull dev. it was already fixed.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00030-4015596898-1girl, two side (...).png, 3MiB, 1280x1792

Anonymous Sat 13 May 2023 18:39:13 No.7373610 Report

Quoted By:

testing
testing...

Anonymous

Anonymous Sat 13 May 2023 18:39:53 No.7373611 Report

Quoted By:

>>7373609
Oh, thank fuck

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765640-2463350838-kill (...).png, 2MiB, 1600x2400

Anonymous Sat 13 May 2023 18:41:14 No.7373612 Report

Quoted By: >>7373624

>>7373315
here you go, hope you like it. let me know if you want different clothing since i want to try more different clothes

Anonymous

Anonymous Sat 13 May 2023 18:41:29 No.7373613 Report

Quoted By: >>7373622

>>7373603
Eh, if you've vram to spare then yeah I guess.
>1600
I discovered the joy of achievement hunting, way funnier than expected and actually a motivation.

>>7373605
>not using additional networks
lol, lmao.
is git pulling safe now in general? im sitting on a month+ old ver

Anonymous

View Same Google ImgOps iqdb SauceNAO 00158-159529597.png, 1MiB, 1536x1024

Anonymous Sat 13 May 2023 18:42:47 No.7373618 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 18:43:16 No.7373620 Report

Quoted By:

>>7373605
a9fed7c3 chads keep winning

Anonymous

View Same Google ImgOps iqdb SauceNAO file.png, 16KiB, 935x192

Anonymous Sat 13 May 2023 18:43:16 No.7373621 Report

Quoted By: >>7373623 >>7373626 >>7373755

Anonymous

Anonymous Sat 13 May 2023 18:43:35 No.7373622 Report

Quoted By: >>7373626

>>7373613
>I discovered the joy of achievement hunting, way funnier than expected and actually a motivation.
I've gone the opposite way, actually. (Besides not paying any more cents to PDX to actually get achievments). I transitioned back into non-ironman.
>is git pulling safe now in general?
I tried to pull and literally every basic feature was broken, so maybe not.

Anonymous

Anonymous Sat 13 May 2023 18:44:34 No.7373623 Report

Quoted By:

>>7373621
KEK

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_tq3ggn.png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 18:46:12 No.7373624 Report

Quoted By: >>7373641

>>7373282
>>7373612
Just letting you know I appreciate thiccjoy

Anonymous

View Same Google ImgOps iqdb SauceNAO 00076-4030752004.png, 1MiB, 1152x1280

Anonymous Sat 13 May 2023 18:46:36 No.7373625 Report

Quoted By: >>7373627

Anonymous

Anonymous Sat 13 May 2023 18:46:53 No.7373626 Report

Quoted By: >>7373636

>>7373621
lol. where the fuck does he even find these loras, the ones I've seen were all properly named

>>7373622
too easy/sandboxy for me that way.
rip pulling

Anonymous

View Same Google ImgOps iqdb SauceNAO file.png, 136KiB, 558x205

Anonymous Sat 13 May 2023 18:48:12 No.7373627 Report

Quoted By: >>7373631 >>7373636 >>7373638 >>7373643

>>7373625
these additional straps..
they always appear, AOM has really poisoned the well, wish someone went back to NAI and do whatever AOM did but better instead of remixing AOM again

Anonymous

Anonymous Sat 13 May 2023 18:48:40 No.7373631 Report

Quoted By: >>7373638 >>7373648

>>7373627
theyre hot tho

Anonymous

Anonymous Sat 13 May 2023 18:49:38 No.7373633 Report

Quoted By:

>>7373591
Naruhodo, try from_below and looking_up maybe idk

Anonymous

Anonymous Sat 13 May 2023 18:51:03 No.7373636 Report

Quoted By: >>7373644

>>7373626
>too easy/sandboxy for me that way.
Well, I always had a skill issue when it comes to EU4. Never cared to do the schizo spreadsheet stuff. I only found out about siege barrages today, lmao.
>rip pulling
Maybe if you don't use img2img or txt2img you can do it??? kek
>>7373627
But AOM is based...

Anonymous

Anonymous Sat 13 May 2023 18:51:37 No.7373638 Report

Quoted By: >>7373648

>>7373631
based

>>7373627
cringe desu

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765620-2648198198-high (...).png, 1MiB, 1280x1440

Anonymous Sat 13 May 2023 18:54:05 No.7373641 Report

Quoted By: >>7373646

>>7373624
huge thanks, i really appreciate that a lot. if you have any killjoy request let me know, depending on request i will try to do it

Anonymous

Anonymous Sat 13 May 2023 18:54:37 No.7373642 Report

Quoted By: >>7373651 >>7373698

>>7373558
>>7373585
Author is suggesting you double the steps with the reference preprocessor, a few good comparisons here https://github.com/Mikubill/sd-webui-controlnet/discussions/1236#discussioncomment-5893225

Anonymous

Anonymous Sat 13 May 2023 18:54:45 No.7373643 Report

Quoted By: >>7373648

>>7373627
interesting wishes
neg "highleg" in the meanwhile

Anonymous

Anonymous Sat 13 May 2023 18:55:35 No.7373644 Report

Quoted By:

>>7373636
>kek
INPAINT GODS RISE UP
>skill issue
no wonder this thread and the internet is what it is, considering that kinda brainlet approach is standard for alot of stuff

Anonymous

View Same Google ImgOps iqdb SauceNAO tmpu4vo6lbx.png, 2MiB, 1536x1024

Anonymous Sat 13 May 2023 18:55:36 No.7373645 Report

Quoted By: >>7373650 >>7373654

Anonymous

Anonymous Sat 13 May 2023 18:56:02 No.7373646 Report

Quoted By: >>7373653 >>7373796

>>7373641
Post-creampied killjoy

Anonymous

Anonymous Sat 13 May 2023 18:56:43 No.7373648 Report

Quoted By:

>>7373643
its not really an issue since I can just lama or inpaint it so its w/e

>>7373638
>>7373631
I agree they're hot but them popping up everywhere kinda ruined it, lost its magic and looks more like an AI glitch to me than anything else at this point

Anonymous

View Same Google ImgOps iqdb SauceNAO 00090-1801115649.png, 1MiB, 1152x1280

Anonymous Sat 13 May 2023 18:58:11 No.7373650 Report

Quoted By: >>7373652

>>7373645
Model? it looks pretty pretty good, can't imagine how much time it took to inpaint.

Anonymous

Anonymous Sat 13 May 2023 18:59:25 No.7373651 Report

Quoted By: >>7373698

>>7373642
Oh, I'll try that then

Anonymous

Anonymous Sat 13 May 2023 18:59:50 No.7373652 Report

Quoted By:

>>7373650
based64 v3, took wayyy too long to inpaint one of the character in

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765603-2803568421-kill (...).png, 1MiB, 1280x2080

Anonymous Sat 13 May 2023 18:59:52 No.7373653 Report

Quoted By:

>>7373646
sounds interesting. might make it just might have to find lora for it

Anonymous

Anonymous Sat 13 May 2023 18:59:57 No.7373654 Report

Quoted By:

>>7373645
Nice inpainting anon

Anonymous

Anonymous Sat 13 May 2023 19:00:57 No.7373655 Report

Quoted By: >>7373658

Hey I’m really new to this forum and wanted to expand my collection please send me anything 219@789@9127

Anonymous

Anonymous Sat 13 May 2023 19:01:33 No.7373656 Report

Quoted By:

sent ;)

Anonymous

Anonymous Sat 13 May 2023 19:02:34 No.7373658 Report

Quoted By:

>>7373655
>no cool handle
this is not a forum post poser

~sent from my blackberry~

Anonymous

Anonymous Sat 13 May 2023 19:03:43 No.7373659 Report

Quoted By: >>7373661 >>7373664

What can I do to backup before updating in case the update is shit?

Anonymous

Anonymous Sat 13 May 2023 19:04:08 No.7373660 Report

Quoted By: >>7373740

>>7373539
>>7372473
catbox please

Anonymous

Anonymous Sat 13 May 2023 19:04:25 No.7373661 Report

Quoted By: >>7373669

>>7373659
There's no need to backup since you can just revert to a previous version

Anonymous

Anonymous Sat 13 May 2023 19:04:45 No.7373664 Report

Quoted By: >>7373669

>>7373659
You can just revert commits, but copy paste the folder if you really want to be sure.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00108-530932787.png, 2MiB, 1152x1280

Anonymous Sat 13 May 2023 19:04:47 No.7373665 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 19:04:52 No.7373666 Report

Quoted By:

https://github.com/Mikubill/sd-webui-controlnet/discussions/1163
>Ignore the color offset problem (!) ... If you can pretend that you have never seen the smaller image, you will forget the color offsets. Just pretend that you have never seen the smaller image, and appreciate the bigger image with great details

Anonymous

Anonymous Sat 13 May 2023 19:06:02 No.7373669 Report

Quoted By: >>7373672

>>7373664
>>7373661
What's the way to revert?

Anonymous

Anonymous Sat 13 May 2023 19:08:25 No.7373672 Report

Quoted By:

>>7373669
git checkout <commit hash>
You'll probably need to stow away your webui-user for that for a second.

Anonymous

Anonymous Sat 13 May 2023 19:11:07 No.7373677 Report

Quoted By: >>7373680

Is the new auto1111 update fixed or is it still breaking everything?

Anonymous

Anonymous Sat 13 May 2023 19:11:46 No.7373680 Report

Quoted By:

>>7373677
Shit's fucked

Anonymous

View Same Google ImgOps iqdb SauceNAO 12586624449.jpg, 914KiB, 2400x3072

Anonymous Sat 13 May 2023 19:18:15 No.7373691 Report

Quoted By: >>7373699

Anonymous

View Same Google ImgOps iqdb SauceNAO xyz_grid-2014-1199692057-1girl, (...).jpg, 508KiB, 2880x2451

Anonymous Sat 13 May 2023 19:23:00 No.7373698 Report

Quoted By:

>>7373642
>>7373651
https://files.catbox.moe/pzd5ab.png
Here are the results in case anyone cares. Based on this sample size of one (1) prompt I think I'll just stick to 20 steps because it doesn't seem like there's a big enough difference to justify the longer generation time

Anonymous

Anonymous Sat 13 May 2023 19:24:45 No.7373699 Report

Quoted By: >>7373707

>>7373691
catbox?

Anonymous

Anonymous Sat 13 May 2023 19:30:15 No.7373707 Report

Quoted By: >>7373710

>>7373699
https://files.catbox.moe/3mpm5c.png
The Rosa lora is my new one, in the same mega as Ilulu, and wamu is one of the two in gayshit list.

Anonymous

Anonymous Sat 13 May 2023 19:32:03 No.7373710 Report

Quoted By:

>>7373707
I'll hunt around for your mega then, thanks

Anonymous

Anonymous Sat 13 May 2023 19:43:06 No.7373729 Report

Quoted By: >>7374074

>>7373554
Some editing was required, well I got sick of variation seed rolling.
https://files.catbox.moe/y7nuig.png

Anonymous

View Same Google ImgOps iqdb SauceNAO 12586624450.jpg, 879KiB, 2400x3072

Anonymous Sat 13 May 2023 19:44:05 No.7373732 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00526-4001239194.png, 842KiB, 640x960

Anonymous Sat 13 May 2023 19:50:45 No.7373740 Report

Quoted By: >>7373966

>>7373660
https://files.catbox.moe/6hopkr.png

the other one is control netted so no real point in sharing it.

if you're interested in the character and not the style then you can get it here: https://mega.nz/folder/IwZDiLZY#SAXoDBZJwkwPdD7OEQgtHA

also is there a way to share direct links to mega subfolders while still giving access to the higher level folder. the gayshit anon has added a couple of mine as separate entries since I direct linked them and that list is already long enough as is

Anonymous

View Same Google ImgOps iqdb SauceNAO 00078-4245541162-nsfw, (extremel (...).png, 1MiB, 1024x1280

Anonymous Sat 13 May 2023 19:52:51 No.7373742 Report

Quoted By: >>7373743 >>7373745 >>7373775 >>7373785 >>7373818 >>7373839

Any good lora combinations you can recommend? I still havent found anything better than as109 and houtengeki (0.6 , 0.6) from the rentry guide

Anonymous

Anonymous Sat 13 May 2023 19:53:07 No.7373743 Report

Quoted By:

>>7373742
mine
:)

Anonymous

Anonymous Sat 13 May 2023 19:53:16 No.7373745 Report

Quoted By:

>>7373742
AWOOOOGA

Anonymous

Anonymous Sat 13 May 2023 20:06:41 No.7373755 Report

Quoted By:

>>7373621
the sheer amount of damage this one youtube video has caused is astounding

Anonymous

Anonymous Sat 13 May 2023 20:12:07 No.7373764 Report

Quoted By: >>7373885

>>7373539
what is multidiffusion?

Anonymous

View Same Google ImgOps iqdb SauceNAO madohexmaniax.png, 433KiB, 512x768

Anonymous Sat 13 May 2023 20:14:31 No.7373769 Report

Quoted By:

well its about time I update my madoka model,

Anonymous

Anonymous Sat 13 May 2023 20:16:39 No.7373775 Report

Quoted By: >>7373785

>>7373742
cat
BOX

Anonymous

Anonymous Sat 13 May 2023 20:25:43 No.7373785 Report

Quoted By: >>7373790

>>7373775
I will give it to you once my riddle>>7373742
has been answered

Anonymous

Anonymous Sat 13 May 2023 20:27:39 No.7373790 Report

Quoted By: >>7373798

>>7373785
my favorite is shuz alexi (0.3, 0.5)

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765645-1857585041-kill (...).png, 2MiB, 1600x2200

Anonymous Sat 13 May 2023 20:33:58 No.7373796 Report

Quoted By: >>7373801

>>7373646
here you go anon. i might do more later

Anonymous

Anonymous Sat 13 May 2023 20:34:43 No.7373798 Report

Quoted By: >>7373808 >>7373809 >>7373827

>>7373790
ty https://files.catbox.moe/ysv1f7.png

Anonymous

Anonymous Sat 13 May 2023 20:35:41 No.7373801 Report

Quoted By: >>7373925

>>7373796
thanks bro

Anonymous

View Same Google ImgOps iqdb SauceNAO 00018-4245541162.png, 1MiB, 1024x1280

Anonymous Sat 13 May 2023 20:40:51 No.7373808 Report

Quoted By:

>>7373798
holy schizo batman

Anonymous

Anonymous Sat 13 May 2023 20:42:02 No.7373809 Report

Quoted By: >>7373846

>>7373798
bruh this dood still using schizo prompts

Anonymous

View Same Google ImgOps iqdb SauceNAO 00252-4251583015.png, 1MiB, 1152x1280

Anonymous Sat 13 May 2023 20:48:43 No.7373817 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 12586624451.jpg, 949KiB, 2304x3072

Anonymous Sat 13 May 2023 20:48:54 No.7373818 Report

Quoted By:

>>7373742
wamudraws + nuclearwasabi gives a style i enjoy. Both with low strength usually

Anonymous

View Same Google ImgOps iqdb SauceNAO image_2023-05-14_005029496.png, 449KiB, 512x768

Anonymous Sat 13 May 2023 20:51:04 No.7373822 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 20:51:28 No.7373824 Report

Quoted By: >>7373836 >>7373840

Is this opll lora a public thing? Some guy shared it on his fanbox. I know there is one floating around out there, just not sure if it's this. I've been liking it but now I have to generate in batches of five because every generation is perfect in its own way.

Anonymous

Anonymous Sat 13 May 2023 20:51:50 No.7373827 Report

Quoted By:

>>7373798
look at this dooood

Anonymous

Anonymous Sat 13 May 2023 20:52:14 No.7373829 Report

Quoted By: >>7373907 >>7373908

>>7373294
>at the cost of quality.
man the stupid shit never ends does it?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00259-444512106.png, 1MiB, 1152x1280

Anonymous Sat 13 May 2023 20:53:58 No.7373834 Report

Quoted By: >>7373897 >>7373927

Anonymous

Anonymous Sat 13 May 2023 20:54:34 No.7373836 Report

Quoted By:

>>7373824
I'm not familiar with it

Anonymous

Anonymous Sat 13 May 2023 20:55:30 No.7373839 Report

Quoted By:

>>7373742
I've always enjoyed combining Raita and Krekkov. One draws very neat and clean, the other dirty, It's like combining sweet and salty.

Anonymous

View Same Google ImgOps iqdb SauceNAO xyz_grid-0003-3322068361.png, 3MiB, 2560x868

Anonymous Sat 13 May 2023 20:56:16 No.7373840 Report

Quoted By: >>7373866

>>7373824
forgot pic

Anonymous

Anonymous Sat 13 May 2023 20:58:40 No.7373846 Report

Quoted By:

>>7373809
w-what about it?

Anonymous

Anonymous Sat 13 May 2023 21:05:53 No.7373864 Report

Quoted By: >>7373884

https://civitai.com/models/7241?modelVersionId=34559
Has anyone tried models from civitai before? How good is this one?

Anonymous

Anonymous Sat 13 May 2023 21:06:16 No.7373866 Report

Quoted By: >>7373876

>>7373840
what fanbox is that?

Anonymous

Anonymous Sat 13 May 2023 21:08:55 No.7373876 Report

Quoted By: >>7373892

>>7373866
hiking, some chink, didn't mention if he baked it or got it from somewhere else

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_dqs93j.jpg, 398KiB, 1536x2304

Anonymous Sat 13 May 2023 21:09:09 No.7373877 Report

Quoted By:

>>7373508
Download the catbox images in this thread and drop them on the "PNG Info" tab of your SD. Use this to see the prompts, models, LoRAs and other configs we use.

Anonymous

Anonymous Sat 13 May 2023 21:10:24 No.7373881 Report

Quoted By: >>7373889 >>7373891

I noticed a lot of people in the /b/ threads use models no one in here even uses. Is based64/65 like an /h/ favored thing or are /b/ threads just that behind on the times?

Anonymous

Anonymous Sat 13 May 2023 21:10:56 No.7373884 Report

Quoted By:

>>7373864
>Has anyone tried models from civitai before?
no
chad.jpg

Anonymous

View Same Google ImgOps iqdb SauceNAO 00210-202619294.png, 3MiB, 1920x1280

Anonymous Sat 13 May 2023 21:11:00 No.7373885 Report

Quoted By:

>>7373764
https://github.com/pkuliyi2015/multidiffusion-upscaler-for-automatic1111

basically generates images in tiles, or run the VAE through tiles, reducing vram consumption.

it also has a feature where specific areas of an image can be given their own prompts so you can gen multiple characters with different LORAs without overlap happening. you probably need controlnet to make it usable for anything other that "girl standing" though

Anonymous

Anonymous Sat 13 May 2023 21:12:36 No.7373889 Report

Quoted By: >>7373893

>>7373881
>Is based64/65 like an /h/ favored thing
Yeah
A lot of the other boards lean more towards SD based models, especially the 3DPD ones

Anonymous

Anonymous Sat 13 May 2023 21:13:22 No.7373891 Report

Quoted By:

>>7373881
>Is based64/65 like an /h/ favored thing
Yeah, sure.
I don't even use it but I think it's an okay base.

Anonymous

View Same Google ImgOps iqdb SauceNAO image_2023-05-14_011322656.png, 3KiB, 272x144

Anonymous Sat 13 May 2023 21:13:31 No.7373892 Report

Quoted By: >>7373896

>>7373876
found him on kemono, he shared the lora recently

Anonymous

Anonymous Sat 13 May 2023 21:13:40 No.7373893 Report

Quoted By: >>7373900

>>7373889
How would you know what a model is based on? Dissecting it after downloading it? So many models look the same so I'm not sure what is what.

Anonymous

Anonymous Sat 13 May 2023 21:15:25 No.7373896 Report

Quoted By: >>7373898 >>7373900

>>7373892
Yeah did you figure out what the rpg thing is? I assume facial expressions but I can't read chink.

Anonymous

Anonymous Sat 13 May 2023 21:16:33 No.7373897 Report

Quoted By:

>>7373834
KATANA WA-...

Anonymous

Anonymous Sat 13 May 2023 21:17:06 No.7373898 Report

Quoted By: >>7373901

>>7373896
judging by trigger words looks like scat

Anonymous

View Same Google ImgOps iqdb SauceNAO 00000-3976151154-(masterpiece, b (...).jpg, 398KiB, 1536x2304

Anonymous Sat 13 May 2023 21:17:24 No.7373899 Report

Quoted By: >>7373920

>>7373508
Download the catbox images in this thread and drop them on the "PNG Info" tab of your SD. Use this to see the prompts, models, LoRAs and other configs we use.

https://files.catbox.moe/dqs93j.jpg

Anonymous

Anonymous Sat 13 May 2023 21:17:40 No.7373900 Report

Quoted By: >>7373906

>>7373893
If someone shares the raw image with its metadata intact (for instance through catbox) you can see the model's hash although since all sorts of random model mixes are becoming increasingly popular this some times feels kind of useless

>>7373896
If the name is of any indication it's like >>>/d/10570341

Anonymous

Anonymous Sat 13 May 2023 21:17:58 No.7373901 Report

Quoted By:

>>7373898
wtf fucking chinks

Anonymous

Anonymous Sat 13 May 2023 21:26:43 No.7373906 Report

Quoted By: >>7373914 >>7373919

>>7373900
For some reason the goo scat doesn't disgust me like regular scat, wonder where it came from feels like it just popped up over night a couple of years ago. Don't find the personality excretion stuff interesting but that's probably cause I don't like mean porn.

Anonymous

Anonymous Sat 13 May 2023 21:28:41 No.7373907 Report

Quoted By: >>7373908 >>7373917 >>7373931

>>7373829
>ignores papers that claim lower batch = better
>ignores test outcomes that show lower batch = better
yeah, it does not. gonna queue a batch 20 next?

Anonymous

Anonymous Sat 13 May 2023 21:29:12 No.7373908 Report

Quoted By: >>7373913 >>7373931

>>7373829
>>7373907
kys fags

Anonymous

Anonymous Sat 13 May 2023 21:29:57 No.7373911 Report

Quoted By: >>7373912 >>7373920

I remember setting up stable diffusion a long time ago but it really sucked with hands and couldn't render anything that wasn't literal mainstream. What's the best up to date guide for a mouth breather to use so that they can start creating their favorite anime girls?

Anonymous

Anonymous Sat 13 May 2023 21:30:28 No.7373912 Report

Quoted By:

>>7373911
>guide
>best
>up to date
lmao

Anonymous

Anonymous Sat 13 May 2023 21:30:47 No.7373913 Report

Quoted By: >>7373915

>>7373908
go do 100 repeats or whatever the youtube faggot tells you to.
training science is important

Anonymous

Anonymous Sat 13 May 2023 21:30:53 No.7373914 Report

Quoted By:

>>7373906
Thats just how fetishes work these days, especially on /d/. Someone has a random idea, it gains traction then fades out of relevance with only a couple of people left truly into it

Anonymous

Anonymous Sat 13 May 2023 21:32:13 No.7373915 Report

Quoted By: >>7373918

>>7373913
kys faggot

Anonymous

Anonymous Sat 13 May 2023 21:33:55 No.7373917 Report

Quoted By: >>7373918 >>7373930

>>7373907
that's weird it, didn't seem to matter that the paper didn't test batch size 1 before, but now they're important again

Anonymous

Anonymous Sat 13 May 2023 21:35:15 No.7373918 Report

Quoted By:

>>7373915
>>7373917
you too

Anonymous

Anonymous Sat 13 May 2023 21:35:26 No.7373919 Report

Quoted By:

>>7373906
I have no idea where it comes from either. The earliest instance I remember of it was a series of doujins by Fan no Hitori. As the other anon said some times these things sort of just pop up every now and then. This started as a sort of off-shoot or subgenre of women getting transformed into onaholes or other sex toys (which ironically has its own general on /d/) which similarly often has overlaps with petrification/transformations etc
I also detest scat but I'll give this a free pass since it's essentially just means to an end and the main focus tends to lie elsewhere anyway.

Anonymous

Anonymous Sat 13 May 2023 21:35:50 No.7373920 Report

Quoted By:

>>7373911
Just go to Automatic1111 webui's github and follow the instructions there to install. Once the webui is installed, download whatever models/LoRAs you want and do this >>7373899

Anonymous

Anonymous Sat 13 May 2023 21:36:48 No.7373923 Report

Quoted By: >>7373926

bros what's the secret to not getting cloudy and lopsided eyes, I feel like I'm becoming the joker here

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765650-461293238-killj (...).png, 2MiB, 2200x1600

Anonymous Sat 13 May 2023 21:37:06 No.7373925 Report

Quoted By: >>7374110

>>7373801
manage to do one more so here you go. now im going to sleep

Anonymous

Anonymous Sat 13 May 2023 21:37:59 No.7373926 Report

Quoted By: >>7373929

>>7373923
Same as always.
Use less shit models, less shit loras, less shit resolutions.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1683830000640105.png, 2MiB, 1080x1280

Anonymous Sat 13 May 2023 21:39:16 No.7373927 Report

Quoted By:

>>7373834

Anonymous

Anonymous Sat 13 May 2023 21:42:35 No.7373929 Report

Quoted By: >>7373934 >>7373935 >>7373937 >>7373940 >>7373943 >>7373950

>>7373926
what's the thread canon not-shit resolution to use bro? I've just been doing the default 512x512 that I've always seen recommended, my [spoiler:lit]laptop[/spoiler:lit] is not that powerful and I think my 4gb vram won't cut it for anything higher?
Would appreciate model and lora recommendations too, I've just been using high rated shit off of civitai
sorry for the newfag questions but all the guides, megas and catboxes are a little hard to navigate

Anonymous

Anonymous Sat 13 May 2023 21:42:38 No.7373930 Report

Quoted By:

>>7373917
what?

Anonymous

Anonymous Sat 13 May 2023 21:42:46 No.7373931 Report

Quoted By:

>>7373907
>>7373908
after these few months it's impressive you're still at it. genuinely hope it's just shitposting, because if not maybe it's an outlet or something for insecurity, idk

Anonymous

Anonymous Sat 13 May 2023 21:47:36 No.7373934 Report

Quoted By: >>7373936 >>7373945 >>7373979

>>7373929
512x768 for brainlets but still better t han 1:1.
640x896 for chad 768 lora users.
Use aom2 mixes: based64/aom2nutmeg/aom2strawberry
Loras? UHhh, check gayshit.
Also use xformers or sdp.

Anonymous

Anonymous Sat 13 May 2023 21:47:42 No.7373935 Report

Quoted By:

>>7373929
512x512 or 768x512 BASE resolution are the most common resolutions people generate at. Hires fix and upscaling is where most of the magic actually happens
As for eyes some times being blurry in spite of this that's some times just a model/lora issue. If that's the case there's not a whole lot you can do about the unaltered generated images, but you can some times make them appear better by inpainting them if you can be bothered with that

Anonymous

Anonymous Sat 13 May 2023 21:48:22 No.7373936 Report

Quoted By: >>7373938

>>7373934
wtf 768 loras work better at that random resolution? is this some math magic

Anonymous

Anonymous Sat 13 May 2023 21:49:21 No.7373937 Report

Quoted By: >>7373979

>>7373929
Depends on what you're genning. I like going 512x768 for cowboy shot or full body. 512x512 is good if you're only going for upper body and maybe if the subject of your image fits a square frame well. If you want better eyes then try prompting for it with stuff like eye focus, detailed eyes, black kohl, very long eyelashes, and maybe some tags that make the image more realistic like lips and photorealistic. What models are you using? Can't go wrong with based64 and AOM2.

Anonymous

Anonymous Sat 13 May 2023 21:50:27 No.7373938 Report

Quoted By:

>>7373936
It's my most common bucket at 768. But you can try the others too.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00002-3976151154-(masterpiece, b (...).png, 752KiB, 536x960

Anonymous Sat 13 May 2023 21:51:41 No.7373940 Report

Quoted By: >>7373950 >>7373979

>>7373929
Use 512x768 with hires fix enabled. 1.5~2x scale and 0.5~0.75 denoising. If it still too high for your card, use 360p as base and 1.5x upscale (this image was made like this). Use --lowvram on your args.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00075-3725574361.png, 547KiB, 512x768

Anonymous Sat 13 May 2023 21:51:48 No.7373941 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00066-2580134192.png, 533KiB, 512x768

Anonymous Sat 13 May 2023 21:52:20 No.7373942 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 21:52:46 No.7373943 Report

Quoted By: >>7373950

>>7373929
btw if you can't upscale like everyone's suggesting try sd upscale script in img2img, uses a lot less vram but gives tiling

Anonymous

Anonymous Sat 13 May 2023 21:53:18 No.7373945 Report

Quoted By: >>7373953

>>7373934
you can also use multidiffusion and its regional prompt to do as highres as you want from the get go, though it may still add undesirable things

Anonymous

View Same Google ImgOps iqdb SauceNAO 1676349188443481.png, 2MiB, 1080x1280

Anonymous Sat 13 May 2023 21:55:50 No.7373947 Report

Quoted By: >>7374655

Don't worry guys I will take over genning Killjoy while the other one is away

Anonymous

Anonymous Sat 13 May 2023 21:57:18 No.7373949 Report

Quoted By: >>7373954

every time I use multidiffusion I get multiple bodies fused together. Is the regional prompt absolutely necessary?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00001-3976151154.0-(masterpiece, (...).png, 2MiB, 1072x1920

Anonymous Sat 13 May 2023 21:57:29 No.7373950 Report

Quoted By:

>>7373929
>>7373940
If you want to upscale a image further, send it to img2img and use the "Ultimate SD Upscale" extension, like >>7373943
said. (last image upscaled by 2x)

Anonymous

View Same Google ImgOps iqdb SauceNAO 00032-3012293989-1girl, two side (...).png, 2MiB, 1280x1792

Anonymous Sat 13 May 2023 21:57:44 No.7373951 Report

Quoted By: >>7373985 >>7373996

Reiq lora. Trained to work at 1.0 with my loras.
https://files.catbox.moe/c6tooz.safetensors
https://rentry.org/gktwz

Anonymous

Anonymous Sat 13 May 2023 21:58:23 No.7373953 Report

Quoted By: >>7373989 >>7373994

>>7373945
how do you upscale with multidiff?

Anonymous

Anonymous Sat 13 May 2023 21:58:36 No.7373954 Report

Quoted By:

>>7373949
yes, refer to their examples. if you don't place the subject unto the background as extension dev tells you to, you're not escaping the problems of tiling that come with highres gens

Anonymous

Anonymous Sat 13 May 2023 22:04:55 No.7373966 Report

Quoted By:

>>7373740
thank you

Anonymous

Anonymous Sat 13 May 2023 22:09:18 No.7373979 Report

Quoted By:

>>7373934
Thanks a lot, had no idea about xformers and it seems like a game changer. Also turned out there was some shit that I had set up completely wrong (sd was mixing python versions up) while I was working that out which I've fixed now.
>>7373937
>>7373940
Thank you. I've been using hassakuHentai and meinaHentai to varying results. I'll give based64 a try.

Anonymous

View Same Google ImgOps iqdb SauceNAO 74391508_p0.jpg, 623KiB, 828x1169

Anonymous Sat 13 May 2023 22:10:41 No.7373984 Report

Quoted By:

Hello. I hope it's alright to request a Desketching of this art of Mai Shiranui and Andy Bogard.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00033-1223435066-1girl, purple h (...).png, 2MiB, 1280x1792

Anonymous Sat 13 May 2023 22:11:13 No.7373985 Report

Quoted By: >>7373996 >>7374170 >>7374173

>>7373951

Anonymous

View Same Google ImgOps iqdb SauceNAO 1.jpg, 456KiB, 1920x3200

Anonymous Sat 13 May 2023 22:13:42 No.7373989 Report

Quoted By: >>7373994

>>7373953
I just use their suggested i2i settings, so low denoise(0.4 and lower). Keep in mind this means it won't gacha-fix your errors, so don't feed it shit. You get back what you give it, only higher res and with clearer lines and details. Mostly worth it for better faces imo

Anonymous

Anonymous Sat 13 May 2023 22:15:21 No.7373994 Report

Quoted By:

>>7373989
>>7373953
oh and yeah, you do want to have regions defining your details with prompt tokens. If you just do a low denoise upscale with no regional proompting, you actually lose details instead of improving them

Anonymous

Anonymous Sat 13 May 2023 22:16:17 No.7373996 Report

Quoted By: >>7374001

>>7373951
>>7373985
that doesn't look like reiq

Anonymous

View Same Google ImgOps iqdb SauceNAO 00136-3762073721.png, 544KiB, 768x512

Anonymous Sat 13 May 2023 22:17:02 No.7373997 Report

Quoted By: >>7374035

Anonymous

Anonymous Sat 13 May 2023 22:18:39 No.7374001 Report

Quoted By: >>7374006 >>7374019

>>7373996
ok

Anonymous

Anonymous Sat 13 May 2023 22:19:49 No.7374006 Report

Quoted By: >>7374015

>>7374001
it's underbaked anon

Anonymous

Anonymous Sat 13 May 2023 22:22:25 No.7374015 Report

Quoted By: >>7374019

>>7374006
ok

Anonymous

Anonymous Sat 13 May 2023 22:25:11 No.7374019 Report

Quoted By: >>7374022 >>7374224

>>7374001
>>7374015
why are the newer bakers here so sensitive when someone tells them their lora has issues?

Anonymous

Anonymous Sat 13 May 2023 22:26:08 No.7374022 Report

Quoted By:

>>7374019
idk

Anonymous

Anonymous Sat 13 May 2023 22:31:42 No.7374029 Report

Quoted By:

God bless Czkawka.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00035-3245200451.png, 1MiB, 960x1280

Anonymous Sat 13 May 2023 22:34:43 No.7374035 Report

Quoted By: >>7374038

>>7373997
How does this make you feel?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00292-4004032781.png, 3MiB, 1920x1088

Anonymous Sat 13 May 2023 22:34:44 No.7374036 Report

Quoted By: >>7374080

Anonymous

View Same Google ImgOps iqdb SauceNAO megusta.png, 24KiB, 372x325

Anonymous Sat 13 May 2023 22:38:48 No.7374038 Report

Quoted By:

>>7374035
chorny

Anonymous

View Same Google ImgOps iqdb SauceNAO 00171-609217045.png, 574KiB, 512x680

Anonymous Sat 13 May 2023 22:51:34 No.7374045 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00434-3582379289.png, 2MiB, 1152x1280

Anonymous Sat 13 May 2023 23:07:06 No.7374051 Report

Quoted By: >>7374054 >>7374056 >>7374067

Anonymous

View Same Google ImgOps iqdb SauceNAO 00174-1258116737-best quality, ( (...).png, 511KiB, 768x768

Anonymous Sat 13 May 2023 23:08:06 No.7374053 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 23:09:07 No.7374054 Report

Quoted By: >>7374081

>>7374051
whoa, lay off the lavender melons kuki

Anonymous

View Same Google ImgOps iqdb SauceNAO 00013-689104647.png, 358KiB, 512x512

Anonymous Sat 13 May 2023 23:13:10 No.7374056 Report

Quoted By: >>7374059 >>7374081

>>7374051
do we have a sister here?

Anonymous

Anonymous Sat 13 May 2023 23:15:14 No.7374059 Report

Quoted By: >>7374078

>>7374056
vae....

NOW!!!!!!

Anonymous

Anonymous Sat 13 May 2023 23:21:52 No.7374063 Report

Quoted By:

>>7373511
i'll have to try it out. i tested shuffle and posted some of the results here the other day, and yeah, shuffle is a game changer for genning variations of a character without needing to download a lora. ofc you'll still want loras and shit to go with it to make it that much better

Anonymous

Anonymous Sat 13 May 2023 23:23:34 No.7374066 Report

Quoted By: >>7374071

>>7373511
How do you use it at the same time as the shuffle preprocessor? Shuffle and Reference-Only are both preprocessors, so you can only pick one
Am I missing something?

Anonymous

View Same Google ImgOps iqdb SauceNAO 02019-3024112129-faux traditiona (...).png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 23:23:48 No.7374067 Report

Quoted By: >>7374081

>>7374051
Kuki my beloved... Has been so long since I gen something with her...

Anonymous

Anonymous Sat 13 May 2023 23:25:27 No.7374071 Report

Quoted By: >>7374076

>>7374066
I meant shuffle controlnet, I clarified in my reply. Use the reference preprocessor in one unit, then shuffle preprocessor and controlnet in another unit (if you don't have more units for some reason go to settings to add more)

Anonymous

Anonymous Sat 13 May 2023 23:30:00 No.7374074 Report

Quoted By:

>>7373729
gracias, and np, i only use catboxes for learning about prompts+models and also for bashing people's prompts together using an extension. step count and sampler info isn't that useful since i just tweak those according to my own taste anyway

Anonymous

Anonymous Sat 13 May 2023 23:32:39 No.7374076 Report

Quoted By: >>7374082

>>7374071
Oh I see, I missed the follow-up reply.
I've never used multiple CNs at once, do I just input the same image in both tabs and pick different preprocessors for both?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00045-689104647.png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 23:33:13 No.7374078 Report

Quoted By: >>7374081 >>7374103

>>7374059
thanks anon, was from my very beginnings. reGenerated it vith manually selected vae.
better?

Anonymous

Anonymous Sat 13 May 2023 23:33:38 No.7374079 Report

Quoted By: >>7374083

Who is using bmaltais UI for LoRA?
Recently i noticed that all new LoRA I am baking is shit.
I'm still using same settings and datasets but have much worse results.
I noticed that they added some parameters like ss_multires_noise_discount,ss_multires_noise_iterations and mnsr gamma.
I haven't changed any of them though.
Reinstalled UI but didn't helped.
My config https://pastebin.com/ER2XEyuE

Anonymous

Anonymous Sat 13 May 2023 23:33:39 No.7374080 Report

Quoted By:

>>7374036
normally i dont point out sameface for ai but its kind of distracting here

Anonymous

View Same Google ImgOps iqdb SauceNAO 00447-3434984097.jpg, 776KiB, 1664x2432

Anonymous Sat 13 May 2023 23:33:42 No.7374081 Report

Quoted By: >>7374089 >>7374103

>>7374054
The LoRA I trained made Kuki have some big booba by default, but specifyng medium or small breasts changes that.

But I don't.

>>7374056
>>7374078
Maybe I should train an NPC sister LoRA (?
And nice, you added VAE

>>7374067
Kuki is love kuki is life

Testing the 1024x1024 LoRA at 832x1216 base res (since it was the most used bucket resolution) and hi-res fix 2x. Just the image takes 20s to be done on a 4090 IoI.

Anonymous

Anonymous Sat 13 May 2023 23:34:10 No.7374082 Report

Quoted By: >>7374115

>>7374076
Yes, usually need to play with the strength (and maybe CFG) since both at 1 is usually a bit much.
Also make sure you're updated because the reference preprocessor got improved a bit ago https://github.com/Mikubill/sd-webui-controlnet/pull/1257

Anonymous

Anonymous Sat 13 May 2023 23:35:34 No.7374083 Report

Quoted By: >>7374097

>>7374079
snr gamma i havent been able to find a human readable discussion of but it's supposed to make things converge better/faster.

multires/pyramid noise is just an alternative to noise offset https://wandb.ai/johnowhitaker/multires_noise/reports/Multi-Resolution-Noise-for-Diffusion-Model-Training--VmlldzozNjYyOTU2

Anonymous

Anonymous Sat 13 May 2023 23:36:02 No.7374085 Report

Quoted By:

crimsoncomics too kino
i hope the lora comes out good

Anonymous

View Same Google ImgOps iqdb SauceNAO 07308-2750532818-((masterpiece)) (...).png, 2MiB, 1024x1536

Anonymous Sat 13 May 2023 23:39:46 No.7374089 Report

Quoted By: >>7374099

>>7374081
Any chance you could share that kuki lora? the latest thing I have about her it's the old embedding which of course lacks a lot of details

Anonymous

View Same Google ImgOps iqdb SauceNAO 00747-3855083438,.jpg, 177KiB, 1024x1536

Anonymous Sat 13 May 2023 23:42:16 No.7374095 Report

Quoted By:

Anonymous

Anonymous Sat 13 May 2023 23:42:23 No.7374096 Report

Quoted By: >>7374098

is there a tool to scan proportions? obvious issues I try to fix, but sometimes it will look vaguely off and I can't tell if fucking around in gimp made it look better or worse.

Anonymous

Anonymous Sat 13 May 2023 23:43:31 No.7374097 Report

Quoted By:

>>7374083
But it doesn't look like something that can fuck LoRA too much.
In my case LoRA just looses too much details. Also, can you share your config?

Anonymous

Anonymous Sat 13 May 2023 23:44:00 No.7374098 Report

Quoted By: >>7374100

>>7374096
You mean like body proportions?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00459-4008930294.png, 4MiB, 2432x1664

Anonymous Sat 13 May 2023 23:46:11 No.7374099 Report

Quoted By: >>7374108 >>7374310

>>7374089
Yeah NP, this epoch should be good for most things. Activation token is kukigenshin. https://mega.nz/folder/E9IihKoJ#FJiWcbpqK_AAMWI8gTJfUQ

And wow, 1024x1024 LoRAs work on 1216x832 works as well.

If you can try it at lower resolution and tell me how it goes please?

Anonymous

Anonymous Sat 13 May 2023 23:46:13 No.7374100 Report

Quoted By: >>7374111

>>7374098
yeah, shot in the dark here.

Anonymous

Anonymous Sat 13 May 2023 23:47:42 No.7374102 Report

Quoted By: >>7374111

>>7373269
why do people keep using this gutted OP that's missing a lot of fucking important links?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00056-3047734416.png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 23:49:16 No.7374103 Report

Quoted By: >>7374106

>>7374078
>>7374081
Well.. much to redo now. But worth it.
Thanks again.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00057-3047734411.png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 23:49:49 No.7374106 Report

Quoted By:

>>7374103

Anonymous

View Same Google ImgOps iqdb SauceNAO 00274-1527648128.png, 2MiB, 1024x1536

Anonymous Sat 13 May 2023 23:49:56 No.7374107 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 07354-603600991-((masterpiece)), (...).png, 2MiB, 1024x1664

Anonymous Sat 13 May 2023 23:50:00 No.7374108 Report

Quoted By: >>7374114

>>7374099
Thanks for sharing
>If you can try it at lower resolution and tell me how it goes please?
Sure i'll gen a few pics

Anonymous

View Same Google ImgOps iqdb SauceNAO 07238-195936564-1girl, solo, (da (...).png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 23:50:22 No.7374110 Report

Quoted By: >>7374114

>hey that tiled vae thing looks cool lets try it out

what the fuck

>>7373925
Very nice

Anonymous

Anonymous Sat 13 May 2023 23:50:56 No.7374111 Report

Quoted By: >>7374116

>>7374100
There are no good universal proportions, but just use a ruler tool in whatever editing software you use.
>>7374102
NAI speedrun is not important, Jose.

Anonymous

Anonymous Sat 13 May 2023 23:51:43 No.7374112 Report

Quoted By:

> start playing darkest dungeon 2
> after a while all horniness is gone
> haven't prompted a thing in a week
Guess guts, blood, death, corpses and suffering don't exactly help with facilitating libido, even if they are stylized and not realistic.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00466-3933546110.jpg, 1010KiB, 1920x2176

Anonymous Sat 13 May 2023 23:53:24 No.7374114 Report

Quoted By: >>7374138 >>7374148

>>7374108
>Sure i'll gen a few pics
Ty anon

>>7374110
Human with small touchs of elfs and very small touchs of an orc

Anonymous

Anonymous Sat 13 May 2023 23:54:02 No.7374115 Report

Quoted By: >>7374121

>>7374082
I tried playing around with it a bit and I'm mostly impressed by how well the reference preprocessor seems to carry styles over. I tried both with and without the shuffle one on top of it and I'm honestly not sure how much of a difference it actually makes. The shuffle one's really volatile by the way, seems like you really need to keep it at low values to not get a warped mess
So far the best part (to me) seems to be how compatible it is with loras, there are barely any artifacts or traces of frying even when I tried using some lower quality loras. Definitely a powerful tool

Anonymous

Anonymous Sat 13 May 2023 23:54:39 No.7374116 Report

Quoted By: >>7374120

>>7374111
>>7374111
>missing voldy's guide
>missing ui alternative
>missing old archive for updates/FAQ
>missing several rentry resources and guides, which while partially outdated, stiull provide useful info
>missing literally anything with colab
stop pushing your retarded minimalism and your shitty 512x512 op images

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_zhvqer.png, 1MiB, 1024x1024

Anonymous Sat 13 May 2023 23:56:06 No.7374119 Report

Quoted By:

Much better

Anonymous

Anonymous Sat 13 May 2023 23:56:11 No.7374120 Report

Quoted By: >>7374123

>>7374116
>literally all outdated shit
>or is on linked wiki
cope

Anonymous

Anonymous Sat 13 May 2023 23:56:18 No.7374121 Report

Quoted By:

>>7374115
>The shuffle one's really volatile by the way, seems like you really need to keep it at low values to not get a warped mess
in my experience if you change the starting control step to something a bit later like 0.1-0.3 that'll help a lot, since then it'll only apply it after the general composition of the image has formed. it's much more sensitive than the reference processor for sure tho.

Anonymous

Anonymous Sat 13 May 2023 23:58:41 No.7374123 Report

Quoted By: >>7374130

>>7374120
how is colab or voldy's guide outdated? and just because you started yesterday and are too stupid to filter useful information doesn't mean we need to deal with your fucking retarded autism "everyone needs to follow my chosen guides" bullshit. op image tells me everything i need to know about the people using that retarded op

Anonymous

Anonymous Sat 13 May 2023 23:58:45 No.7374126 Report

Quoted By: >>7374128 >>7374169

Anybody else get some fried eyes when trying to do a sideways position? I'm wondering if the models are just expecting upright positions

Anonymous

Anonymous Sat 13 May 2023 23:59:35 No.7374128 Report

Quoted By: >>7374200

>>7374126
Not sure what you mean but upside down fucks up faces bad because the model/lora wasn't trained on upside down faces.

Anonymous

Anonymous Sun 14 May 2023 00:00:14 No.7374130 Report

Quoted By: >>7374134

>>7374123
>how is colab
it's in the op
>voldy's guide
outdated retard shit also superseded by his own op link
cope harder

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_9jlnfm.png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 00:01:42 No.7374132 Report

Quoted By: >>7374216 >>7374221

Dare you enter the garden of the elves?

Anonymous

Anonymous Sun 14 May 2023 00:03:07 No.7374134 Report

Quoted By: >>7374139

>>7374130
>cope harder
parasocial twitch nigger zoomers are insufferable and can't actually say anything with any substance lol

Anonymous

View Same Google ImgOps iqdb SauceNAO 43589-3385143736.0-solo, kukigen (...).png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 00:03:58 No.7374138 Report

Quoted By: >>7374162

>>7374114
Quick test with a simple prompt, seems fine to me

Anonymous

Anonymous Sun 14 May 2023 00:04:00 No.7374139 Report

Quoted By: >>7374144

>>7374134
>parasocial
fuck off zoomer
and cope hardest
or bake the next op instead of waiting 2h like a bitch

Anonymous

View Same Google ImgOps iqdb SauceNAO 00470-1759858658.jpg, 2MiB, 4096x4096

Anonymous Sun 14 May 2023 00:04:55 No.7374142 Report

Quoted By: >>7374153

wtf tiled VAE let me do a 4096x4096 image with hi-res fix IoI, my 4090 used 500W and it took 4 min to generate, not recommended since it can produce some artifacts in the image.

gonna try 5120*5120 anyways.

Anonymous

Anonymous Sun 14 May 2023 00:05:16 No.7374144 Report

Quoted By:

>>7374139
>no you
lmao

Anonymous

Anonymous Sun 14 May 2023 00:10:22 No.7374148 Report

Quoted By: >>7374162

>>7374114
Tiamat lora? Can I have it?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00613-9886406-1girl, long hair, (...).png, 870KiB, 640x896

Anonymous Sun 14 May 2023 00:10:59 No.7374149 Report

Quoted By: >>7374173

My pussy's gone crazy!

Anonymous

Anonymous Sun 14 May 2023 00:13:12 No.7374153 Report

Quoted By:

>>7374142
Jelly. I would if I could.

Anonymous

Anonymous Sun 14 May 2023 00:14:40 No.7374155 Report

Quoted By: >>7374157

>cosine with restarts at anything more than 1 fucks up my results
?

Anonymous

Anonymous Sun 14 May 2023 00:15:19 No.7374157 Report

Quoted By: >>7374158

>>7374155
Makes sense if you're unlucky.

Anonymous

Anonymous Sun 14 May 2023 00:16:25 No.7374158 Report

Quoted By:

>>7374157
i am unlucky in love

Anonymous

View Same Google ImgOps iqdb SauceNAO 2023-05-13_19-12-28.png, 1MiB, 512x768

Anonymous Sun 14 May 2023 00:18:57 No.7374160 Report

Quoted By: >>7374164 >>7374234 >>7374294

Bros, what's a good 90s style lora that isn't Bluethebone or Urushihara or some other modern rendition of it by a current artist? Or do I have to screen cap my old H-animes and make it myself? I'm burnt out on baking and just want to proompt.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00471-1957814592.jpg, 3MiB, 5120x5120

Anonymous Sun 14 May 2023 00:19:15 No.7374162 Report

Quoted By: >>7374165 >>7374167 >>7374190

>>7374138
Nice, thanks anon for testing. Good to know that 1024x1024 LoRAs work at 512x512 base res as well.

Also tested 5120x5120 with hi-res (upscaling from 1024x1024 with 5x multiplier) and tiled VAE let me do it as well, damn. Took 6+ minutes for one image. Prob is the limit since it was using 23-24GB of VRAM.

>>7374148
Which GPU do you have?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00621-479994495-1girl, long hair (...).png, 869KiB, 640x896

Anonymous Sun 14 May 2023 00:20:31 No.7374164 Report

Quoted By: >>7374173

>>7374160
>90s style lora
just prompt it broski

Anonymous

Anonymous Sun 14 May 2023 00:20:33 No.7374165 Report

Quoted By: >>7374174

>>7374162
ma'am get out of the outhouse

Anonymous

Anonymous Sun 14 May 2023 00:21:45 No.7374167 Report

Quoted By: >>7374174

>>7374162
A 1070, why?

Anonymous

Anonymous Sun 14 May 2023 00:24:21 No.7374169 Report

Quoted By: >>7374200

>>7374126
>I'm wondering if the models are just expecting upright positions
Oh they definitely are. I once tried ControlNeting a character turned upside down and the end result looked more like a totem pole with multiple torsos and faces. Depending on how sideways the position you're going for is I imagine you might need a lora. Some times the 'from side' prompt helps but if it's anything beyond that then I'm not really sure what else to do.

Anonymous

Anonymous Sun 14 May 2023 00:25:39 No.7374170 Report

Quoted By:

>>7373985
Speaking of pokemom, anyone wanna give a shot with Terufuu art style? I'm surprised nobody did it yet so far.

Anonymous

Anonymous Sun 14 May 2023 00:27:41 No.7374173 Report

Quoted By: >>7374176

>>7373985
>>7374164
>>7374149

Speaking of pokemom, anyone wanna give a shot with Terufuu art style? I'm surprised nobody did it yet so far.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00472-421471834.png, 4MiB, 1664x2432

Anonymous Sun 14 May 2023 00:28:40 No.7374174 Report

Quoted By: >>7374177

>>7374167
Asking for the VRAM, 8GB maybe can do 3072x3072? With tiled VAE at least. Or 2560x2560.

>>7374165
Gacha images sometimes produce random things IoI

buff ayaka because why not

Anonymous

View Same Google ImgOps iqdb SauceNAO 00632-4211116171-1girl, long hai (...).png, 756KiB, 640x896

Anonymous Sun 14 May 2023 00:29:36 No.7374176 Report

Quoted By: >>7374179 >>7374184 >>7374185

You know, it's kinda sad that promptsmithing for different styles and such died instantly after we moved on from walled garden NAI.
>>7374173
Who's he?

Anonymous

Anonymous Sun 14 May 2023 00:29:50 No.7374177 Report

Quoted By:

>>7374174
I think you quoted the wrong anon. I'm the anon asking for the tiamat lora

Anonymous

Anonymous Sun 14 May 2023 00:31:04 No.7374179 Report

Quoted By: >>7374186

>>7374176
It still exists in the form of combining artstyle loras
I find it quite fun to see how various artists mix together

Anonymous

View Same Google ImgOps iqdb SauceNAO 1669987235341216.png, 2MiB, 1080x1280

Anonymous Sun 14 May 2023 00:33:16 No.7374183 Report

Quoted By: >>7374655

Anonymous

Anonymous Sun 14 May 2023 00:33:19 No.7374184 Report

Quoted By: >>7374187

>>7374176
aye things got real soulless

Anonymous

Anonymous Sun 14 May 2023 00:33:24 No.7374185 Report

Quoted By: >>7374187

>>7374176
Look up Terufuu on Danbooru. His stuff is pretty damn good.

Anonymous

Anonymous Sun 14 May 2023 00:33:26 No.7374186 Report

Quoted By:

>>7374179
True, but there's no exploration when you know what's gonna happen.
Trying a billion different vaguely related words to see how they react was interesting.

Anonymous

Anonymous Sun 14 May 2023 00:35:07 No.7374187 Report

Quoted By: >>7374193 >>7374196

>>7374184
Industrial is a good term for it. Though I'll try not to look at NAI with rose tinted glasses like it was any good.
>>7374185
Oh, him. I remember wanking to his stuff a decade ago. Can't say I like the style though.

Anonymous

Anonymous Sun 14 May 2023 00:36:16 No.7374189 Report

Quoted By: >>7374213

Anyone else like their creations too much to beat off to them

I finally understand artists

Anonymous

View Same Google ImgOps iqdb SauceNAO 43626-565830536.0-solo, kukigens (...).png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 00:37:01 No.7374190 Report

Quoted By: >>7374191

>>7374162
Did a few more tests but they aren't /h/ at all so have catboxes instead >https://files.catbox.moe/cysz5k.png
>https://files.catbox.moe/4a3u9g.png
>https://files.catbox.moe/7nd64w.png
And thanks again for sharing, works pretty well

Anonymous

View Same Google ImgOps iqdb SauceNAO 00477-1184568308.jpg, 815KiB, 1664x2432

Anonymous Sun 14 May 2023 00:41:09 No.7374191 Report

Quoted By: >>7374195

>>7374190
Np, glad you find it great. 1024x1024 LoRAs training paid off. At total, it took 38 hours to train the ~200 LoRAs, on 2x4090 :(

And yep, I make my LoRAs to be able to do SFW and NSFW as well.

Anonymous

Anonymous Sun 14 May 2023 00:44:33 No.7374193 Report

Quoted By: >>7374204

>>7374187
It was my favorite style in old time. I've been wanting to see how Terufuu would mesh with other artist's style.

If anyone see artist requests from anons, would be fantastic if you say Terufuu.

Anonymous

View Same Google ImgOps iqdb SauceNAO 01051-162358911.png, 1MiB, 1040x1560

Anonymous Sun 14 May 2023 00:44:50 No.7374194 Report

Quoted By:

i love vex

Anonymous

Anonymous Sun 14 May 2023 00:45:11 No.7374195 Report

Quoted By: >>7374201

>>7374191
tell us your settings and insights richboi

Anonymous

Anonymous Sun 14 May 2023 00:46:31 No.7374196 Report

Quoted By: >>7374204

>>7374187
>remember wanking to his stuff a decade ago
It's honestly fucked up how many times I've gathered images for datasets and I randomly come across an image I recognise from 10 years ago. I'm really not sure how to feel about it honestly

Anonymous

Anonymous Sun 14 May 2023 00:53:57 No.7374200 Report

Quoted By: >>7374208

>>7374128
>>7374169
man maybe I should just rotate the image 90 degrees when I inpaint/img2img then,
experimenting with this idea now

Anonymous

View Same Google ImgOps iqdb SauceNAO 00481-1064885812.png, 2MiB, 1536x1920

Anonymous Sun 14 May 2023 00:53:57 No.7374201 Report

Quoted By: >>7374218

>>7374195
You can find an example of what setting I use here, and a dataset. https://mega.nz/folder/w0JySCaZ#lSDgRTWSXqRp71eW7WD17w

I use .ps1 (powershell) instead of easyscripts or guis, so no jsons, but I guess you could translate it.

You also have to change the min and max bucket size (I used 512 min and 2048 max)

1024x1024 can get really small details that 768x768 LoRAs can not.

Each LoRA. if using 6 epochs, took 25-30 min. 4 epochs take 20 min or near that.

The most common bucket was 832*1216.

Not sure what more it could be said, you need 24GB of VRAM. I suggest upgrading to torch2.0/2.1+cu118 on the kohya scripts, since it uses less VRAM than torch 1.12 that it comes by default.

Oh and I prefer to use 8bit adamW vs dadaptation. Matter of preference I guess.

Anonymous

Anonymous Sun 14 May 2023 00:54:57 No.7374202 Report

Quoted By: >>7374203 >>7374205 >>7374206

>Too many futa
>Add futanari embed to negatives
>Girls who were formerly stroking their dicks are now using wand vibrators instead
Just putting (futa, futanari:1.7) in the negatives didn't seem to do anything, how do I het less dickgirls while promting men and women in the same image?

Anonymous

Anonymous Sun 14 May 2023 00:58:23 No.7374203 Report

Quoted By:

>>7374202
i guess ask /d/ for advice on prompting futa and then do the opposite of that

Anonymous

Anonymous Sun 14 May 2023 00:58:47 No.7374204 Report

Quoted By: >>7374209

>>7374193
>It was my favorite style in old time
Hmm, I don't remember that many old ones. ZONE was bae. Maybe Edithemad, though I disliked his style even back then.
I definitely remember Terufuu though. Maybe I'll bake him tomorrow just for you and old times' sake.
>>7374196
Hey, tastes can stay and good artists stay good artists.
And it makes me appreciate what we have now, with the hundreds of good artists and now AI. Back then it was a different time. Jerking it to Pokemon and Mortal Kombat chicks just because it was big enough to have some big names (probably from one of those paid studios) working on it, lol!
>I'm really not sure how to feel about it honestly
Tell me more. And which images.

Anonymous

Anonymous Sun 14 May 2023 00:58:58 No.7374205 Report

Quoted By:

>>7374202
try other words like dickgirl, shemale, hermaphrodite, etc. If you want to make the male non-black prompting "nigger" works which is kind of hilarious

Anonymous

Anonymous Sun 14 May 2023 00:59:25 No.7374206 Report

Quoted By:

>>7374202
move your men tokens higher in the prompt and remove solo if you are still using it

Anonymous

Anonymous Sun 14 May 2023 01:00:05 No.7374208 Report

Quoted By:

>>7374200
motherfucker... it actually works to some degree. Last time I kept getting absolute splotches, this time it's round with an actual iris. Not perfect but a huge improvement.

Anonymous

Anonymous Sun 14 May 2023 01:00:53 No.7374209 Report

Quoted By: >>7374210

>>7374204
> Maybe I'll bake him tomorrow just for you and old times' sake.

You're awesome. Take your time. Thanks.

Anonymous

Anonymous Sun 14 May 2023 01:02:22 No.7374210 Report

Quoted By:

>>7374209
Yeah, sure. I wonder what it'll look like.
And ouh, now I remember. Lolita Channel was a classic back then and that style is baked into my mind.

Anonymous

Anonymous Sun 14 May 2023 01:06:16 No.7374213 Report

Quoted By: >>7374214

>>7374189
Do you mean like starting to think of your oc's as your children?

Anonymous

Anonymous Sun 14 May 2023 01:07:20 No.7374214 Report

Quoted By:

>>7374213
I wouldnt go that far

Anonymous

Anonymous Sun 14 May 2023 01:13:01 No.7374215 Report

Quoted By:

Fuck this summer is going to kill me. Am I generating more heat per pixel when I jack up the resolution? Also how is power efficiency effected by high/med/low vram?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00131.png, 2MiB, 1152x1728

Anonymous Sun 14 May 2023 01:14:10 No.7374216 Report

Quoted By: >>7374221 >>7374225

>>7374132
I do.

Also very nice, what style loras did you use and at which strength?

Anonymous

Anonymous Sun 14 May 2023 01:17:21 No.7374218 Report

Quoted By:

>>7374201
>1024x1024 can get really small details that 768x768 LoRAs can not.

do you have any examples of this

Anonymous

Anonymous Sun 14 May 2023 01:17:25 No.7374219 Report

Quoted By: >>7374222

>forget to put anything in the negative prompt
>everything besides details is much better
fucking black box bros...

Anonymous

View Same Google ImgOps iqdb SauceNAO 00132.png, 3MiB, 1344x2016

Anonymous Sun 14 May 2023 01:19:12 No.7374221 Report

Quoted By:

>>7374216
>>7374132

Anonymous

Anonymous Sun 14 May 2023 01:19:26 No.7374222 Report

Quoted By: >>7374227

>>7374219
>lower value of (low quality, worst quality:1.4) in negatives
>result looks more like a drawing and ends up looking better

Anonymous

Anonymous Sun 14 May 2023 01:19:26 No.7374223 Report

Quoted By:

>have to open sd, close it and open again to get normal results otherwise a single launch gives me dog shit for some reason

Anonymous

View Same Google ImgOps iqdb SauceNAO 00051-844993938.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 01:19:51 No.7374224 Report

Quoted By:

>>7374019
Because they don't post to receive criticism, they post to receive attention and validation. Instead of being motivated to better themselves when they don't get the reception they were looking for, they dismiss any negative feedback as a personal attack (fucking hilarious on an anonymous imageboard) and become defensive. It's a symptom of social media addiction and their own immaturity.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_looog7.png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 01:20:50 No.7374225 Report

Quoted By: >>7374228

>>7374216
Thanks, took that one anons suggestion and mixed as109 and ishikei both at .5 strength

Nice voluptuous elf

Anonymous

Anonymous Sun 14 May 2023 01:22:07 No.7374227 Report

Quoted By:

>>7374222
Ok, anon, that was always known.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_e959u5.png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 01:23:15 No.7374228 Report

Quoted By:

>>7374225

Anonymous

View Same Google ImgOps iqdb SauceNAO 00372-2773678864.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 01:23:52 No.7374230 Report

Quoted By:

Cookie. Had a lot of trouble with this one. I regret the missing foot but I don't want to spend more time on this gen.

Anonymous

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

Anonymous Sun 14 May 2023 01:25:00 No.7374232 Report

Quoted By: >>7374235 >>7374242

Anonymous

Anonymous Sun 14 May 2023 01:25:39 No.7374234 Report

Quoted By: >>7374294

>>7374160
just type in 90s anime style bro...

Anonymous

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

Anonymous Sun 14 May 2023 01:26:11 No.7374235 Report

Quoted By: >>7374242

>>7374232

Anonymous

Anonymous Sun 14 May 2023 01:27:33 No.7374238 Report

Quoted By:

so i tried out the fancy tile upscale thing and it's kinda neat

Anonymous

View Same Google ImgOps iqdb SauceNAO 00489-1309435632.png, 3MiB, 1536x1792

Anonymous Sun 14 May 2023 01:27:57 No.7374239 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 01:30:17 No.7374242 Report

Quoted By:

>>7374235
>>7374232
Nice. Catbox?

I keep seeing tiny pom-poms in some pics.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_qrfzcc.png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 01:30:47 No.7374243 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 01:32:12 No.7374244 Report

Quoted By: >>7374247 >>7374248

4090bros, assuming I'm starting from a clean windows install, what do I need to optimize my performance?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00088-908560877.png, 2MiB, 1024x1408

Anonymous Sun 14 May 2023 01:32:19 No.7374245 Report

Quoted By:

i fucking love ai

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_x66e05.png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 01:32:55 No.7374246 Report

Quoted By:

Overflowing creampies is the best

Anonymous

Anonymous Sun 14 May 2023 01:34:48 No.7374247 Report

Quoted By: >>7374257

>>7374244
two 4090s

Anonymous

Anonymous Sun 14 May 2023 01:36:27 No.7374248 Report

Quoted By: >>7374251

>>7374244
undervolt setup (the thread will not help you and will shit on you if you ask), the proper torch/cuda/python setup and possibly a few other tweaks. the git might know.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00059-2437427401.png, 520KiB, 512x768

Anonymous Sun 14 May 2023 01:38:18 No.7374251 Report

Quoted By: >>7374254 >>7374259

>>7374248
i never figured out how to undervolt

Anonymous

Anonymous Sun 14 May 2023 01:40:25 No.7374254 Report

Quoted By: >>7374266

>>7374251
there are some tutorials on youtube, they aren't the greatest and eventually you just kind of have to guess at what your curve should be

Anonymous

Anonymous Sun 14 May 2023 01:44:57 No.7374257 Report

Quoted By:

>>7374247
it was limit one per household :'(

Anonymous

Anonymous Sun 14 May 2023 01:46:16 No.7374258 Report

Quoted By: >>7374264

guess I gotta update since it keeps giving me the same dogshit results no matter the model. fucking annoyed

Anonymous

Anonymous Sun 14 May 2023 01:46:39 No.7374259 Report

Quoted By: >>7374263

>>7374251
literally just download msi afterburner and click on the buttons that limit power usage

Anonymous

Anonymous Sun 14 May 2023 01:47:37 No.7374263 Report

Quoted By: >>7374265

>>7374259
that isn't actually the same thing (though it might help and no one's really explained the difference well enough)

Anonymous

Anonymous Sun 14 May 2023 01:48:31 No.7374264 Report

Quoted By:

>>7374258
What are you trying to do?

Anonymous

Anonymous Sun 14 May 2023 01:48:52 No.7374265 Report

Quoted By:

>>7374263
it has actual undervolting too and fan curve changing and most of the other features that more enthusiast programs have

Anonymous

View Same Google ImgOps iqdb SauceNAO 00064-2857661771.png, 385KiB, 512x768

Anonymous Sun 14 May 2023 01:49:27 No.7374266 Report

Quoted By:

>>7374254
yeah but thats going to be a whole day ,it always ends up being for some reason or another, something somewhere goes wrong and i end up having to spend hours troubleshooting.
i want to enjoy my weekends now that i cant neet

Anonymous

View Same Google ImgOps iqdb SauceNAO 00006-1536956543.png, 1009KiB, 712x1000

Anonymous Sun 14 May 2023 01:50:33 No.7374267 Report

Quoted By: >>7374268

I wish I could get this to run off my GPU instead of my CPU. This took 4m 45secs.

Anonymous

Anonymous Sun 14 May 2023 01:51:36 No.7374268 Report

Quoted By: >>7374273

>>7374267
amd gpu?

Anonymous

Anonymous Sun 14 May 2023 01:53:09 No.7374270 Report

Quoted By:

Is there a gcfmug lora?

Anonymous

Anonymous Sun 14 May 2023 01:55:02 No.7374273 Report

Quoted By:

>>7374268
Yeah.

Anonymous

View Same Google ImgOps iqdb SauceNAO 2023-05-13_20-45-04.png, 36KiB, 247x143

Anonymous Sun 14 May 2023 01:55:09 No.7374274 Report

Quoted By: >>7374285

oh shit

Anonymous

Anonymous Sun 14 May 2023 01:55:21 No.7374275 Report

Quoted By:

Another new extension. I think this is some kind of image editor similar to instructpix2pix but more flexible. https://github.com/ashen-sensored/sd_webui_masactrl

Anonymous

View Same Google ImgOps iqdb SauceNAO 00516-3936798226.png, 2MiB, 1280x1408

Anonymous Sun 14 May 2023 01:57:45 No.7374280 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 02:02:06 No.7374285 Report

Quoted By: >>7374287

>>7374274
which O-type star are you genning on anon?

Anonymous

Anonymous Sun 14 May 2023 02:04:15 No.7374287 Report

Quoted By:

>>7374285
So this is the power of the dyson sphere...

SirOriole

View Same Google ImgOps iqdb SauceNAO 00013-641852249.jpg, 374KiB, 2560x2560

SirOriole Sun 14 May 2023 02:04:51 No.7374288 Report

Quoted By: >>7374290

Anonymous

Anonymous Sun 14 May 2023 02:07:28 No.7374290 Report

Quoted By:

>>7374288
Hi there!

You seem to have made a bit of a mistake in your post. Luckily, the users of 4chan are always willing to help you clear this problem right up! You appear to have used a tripcode when posting, but your identity has nothing at all to do with the conversation! Whoops! You should always remember to stop using your tripcode when the thread it was used for is gone, unless another one is started! Posting with a tripcode when it isn't necessary is poor form. You should always try to post anonymously, unless your identity is absolutely vital to the post that you're making!

Now, there's no need to thank me - I'm just doing my bit to help you get used to the anonymous image-board culture!

Anonymous

View Same Google ImgOps iqdb SauceNAO 00524-2578205110.png, 2MiB, 1280x1408

Anonymous Sun 14 May 2023 02:09:36 No.7374292 Report

Quoted By: >>7374580

Anonymous

View Same Google ImgOps iqdb SauceNAO 00056-2943897986.png, 507KiB, 512x768

Anonymous Sun 14 May 2023 02:11:25 No.7374294 Report

Quoted By:

>>7374234
>>7374160
nta, but while that helps, its not really enough a lot of the time. there's a few loras out there that might help, but ive not had the time to do proper experimentation with them. I was using dh_classicanime as a model for a while, and am still using some mixes with that. i think the guy renamed it on civitai though. dont let the furry crap fool you, its not really a furry model. (then again, ive never tried to gen furry filth), though if i knew how to prune, i'd just remove the furry stuff from it. most of my stuff is a mix of Urushihara, Akemi Takada, and a dash of Zankuro, with occasional bluethebone, or a couple other people. a proper 80s/90s hentai screenshot lora would be kinda awesome though if you make a good one.

Anonymous

Anonymous Sun 14 May 2023 02:23:21 No.7374303 Report

Quoted By: >>7374313 >>7374880

>it's been 30 years since NAI leak...

Anonymous

View Same Google ImgOps iqdb SauceNAO 00538-3446344744.png, 2MiB, 1280x1408

Anonymous Sun 14 May 2023 02:29:14 No.7374306 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 02:30:27 No.7374310 Report

Quoted By: >>7374317

>>7374099
catbox please.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00009-1693341958.png, 487KiB, 512x728

Anonymous Sun 14 May 2023 02:32:24 No.7374312 Report

Quoted By:

At least under Linux, the overhead is lower, so the CPU can churn them out must faster, That 4m render would have taken 10 minutes under Windows.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00002-3027234894-masterpiece, be (...).png, 363KiB, 512x704

Anonymous Sun 14 May 2023 02:33:16 No.7374313 Report

Quoted By: >>7374880

>>7374303
ah, the NAI era...kids these days don't know how good they have it...

Anonymous

View Same Google ImgOps iqdb SauceNAO 00548-1239010555.png, 2MiB, 1280x1408

Anonymous Sun 14 May 2023 02:39:38 No.7374317 Report

Quoted By: >>7374321

>>7374310
https://files.catbox.moe/zqaxxa.png

Anonymous

Anonymous Sun 14 May 2023 02:40:11 No.7374318 Report

Quoted By:

I was down for the day, Can I update shit yet or is it still fucking fucked

Anonymous

Anonymous Sun 14 May 2023 02:43:56 No.7374321 Report

Quoted By: >>7374330

>>7374317
That is massive. Did you have to upscale that? I like the style, too. lora, model, or prompt to make it have that nice coloring? Sorry. Am new.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00010-2294883908.png, 564KiB, 512x728

Anonymous Sun 14 May 2023 02:45:36 No.7374324 Report

Quoted By:

This came out unexpectantly cute.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00561-781660553.png, 2MiB, 1160x1696

Anonymous Sun 14 May 2023 02:56:39 No.7374330 Report

Quoted By: >>7374338

>>7374321
You can download the image from the link and get all the info (prompt, model, etc) using the png info on the webui.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00013-2474061864.png, 546KiB, 512x728

Anonymous Sun 14 May 2023 03:05:02 No.7374338 Report

Quoted By:

>>7374330
Thanks for foonspeeding me, anon.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00579-3960263530.png, 2MiB, 1304x1408

Anonymous Sun 14 May 2023 03:15:06 No.7374348 Report

Quoted By: >>7374354 >>7374371 >>7374454

Help anons, I can't stop making buff versions of any character.

Anonymous

Anonymous Sun 14 May 2023 03:20:06 No.7374354 Report

Quoted By: >>7374375 >>7374454

>>7374348
neuron activation

Anonymous

View Same Google ImgOps iqdb SauceNAO 03365-28732085 (1).png, 575KiB, 512x768

Anonymous Sun 14 May 2023 03:23:44 No.7374358 Report

Quoted By: >>7374424

what do you suppose is going on here?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00017-3730406164.png, 463KiB, 512x728

Anonymous Sun 14 May 2023 03:32:57 No.7374371 Report

Quoted By:

>>7374348
>petite
>short stack
>perky tits
>neuron activation
>"you're a louse!" expression
I like your style, anon

Anonymous

View Same Google ImgOps iqdb SauceNAO 00607-4075232193.png, 2MiB, 1280x1408

Anonymous Sun 14 May 2023 03:36:05 No.7374375 Report

Quoted By: >>7374454

>>7374354
Basically every time I make these.

Anonymous

View Same Google ImgOps iqdb SauceNAO tmp330ed9hm.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 03:40:23 No.7374381 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00613-3475884026.png, 2MiB, 1280x1408

Anonymous Sun 14 May 2023 03:59:16 No.7374403 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 04:12:28 No.7374424 Report

Quoted By: >>7374930

>>7374358
How did you get this, did you put a random hentai manga image into the img2img?

Anonymous

Anonymous Sun 14 May 2023 04:18:13 No.7374433 Report

Quoted By: >>7374439 >>7374474 >>7374500 >>7374503

HOW HARD IS IT TO PREVENT AUTO GENERATION IF THERE IS A SEED SET AHHHHHHHHH

Anonymous

Anonymous Sun 14 May 2023 04:25:12 No.7374439 Report

Quoted By: >>7374474

>>7374433
You mean disabling generate forever when a seed is manually set? I could try making a PR for that

Anonymous

Anonymous Sun 14 May 2023 04:32:03 No.7374442 Report

Quoted By: >>7374783

>>7373300
Catbox?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1677722347797641.png, 3MiB, 1280x1792

Anonymous Sun 14 May 2023 04:45:38 No.7374454 Report

Quoted By:

>>7374348
>>7374354
>>7374375
My brothers.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00053-892636804.png, 1MiB, 1536x1024

Anonymous Sun 14 May 2023 04:47:22 No.7374458 Report

Quoted By: >>7374460

Anonymous

Anonymous Sun 14 May 2023 04:48:22 No.7374460 Report

Quoted By: >>7374468

>>7374458
catbox?

I love when the AI manages to generate decent looking genitalia

Anonymous

View Same Google ImgOps iqdb SauceNAO 00227-2422719180.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 04:56:18 No.7374466 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 04:56:23 No.7374468 Report

Quoted By: >>7374541

>>7374460
It's an inpaint though.
https://files.catbox.moe/l0pxp2.png
It's pretty easy to get these results. I mask off the area, enter "spread pussy" into the prompt with the style LoRA(s) and use Euler a as the sampler at 0.2- 0.7 denoise. I mostly use 0.08 - 0.2 denoise for faces and 0.4 for digits and genitals.0.6 and above is for gacha.
The incase LoRA is a WIP that draenai anon is baking. He randomly linked it in the threads a few days ago if you want it.
https://files.catbox.moe/l0pxp2.png

Anonymous

Anonymous Sun 14 May 2023 05:03:17 No.7374474 Report

Quoted By: >>7374880

>>7374433
>>7374439
done https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/10364

Anonymous

Anonymous Sun 14 May 2023 05:20:30 No.7374484 Report

Quoted By: >>7374488

So many new extensions coming out there's barely any time to experiment with them until a new one comes out..
Future is now.

Anonymous

Anonymous Sun 14 May 2023 05:26:40 No.7374488 Report

Quoted By: >>7374501

>>7374484
>So many new extensions coming out
like
?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00016-1454635304.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 05:29:14 No.7374490 Report

Quoted By: >>7374598

Anonymous

View Same Google ImgOps iqdb SauceNAO 00173-3123960119.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 05:32:56 No.7374495 Report

Quoted By: >>7374497 >>7374499

https://mega.nz/folder/ctl1FYoK#BYlpywutnH4psbEcxsbLLg
added drcockula

Anonymous

View Same Google ImgOps iqdb SauceNAO 00013-395118629.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 05:33:28 No.7374497 Report

Quoted By:

>>7374495

Anonymous

View Same Google ImgOps iqdb SauceNAO 00012-2120064452.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 05:35:20 No.7374498 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 05:35:34 No.7374499 Report

Quoted By: >>7374505

>>7374495
Would you be able to do a gcfmug style?

Anonymous

Anonymous Sun 14 May 2023 05:35:38 No.7374500 Report

Quoted By: >>7374511

>>7374433
lmfao, voldy essentially just said you're retarded https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/10364#issuecomment-1546812125

Anonymous

Anonymous Sun 14 May 2023 05:38:22 No.7374501 Report

Quoted By:

>>7374488
>he doesn't know

Anonymous

Anonymous Sun 14 May 2023 05:41:55 No.7374503 Report

Quoted By:

>>7374433
based retard

Anonymous

View Same Google ImgOps iqdb SauceNAO 00048-1276469909.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 05:44:06 No.7374505 Report

Quoted By: >>7374507

>>7374499
not really interested in futa or furry but if you have other suggestions i wouldn't mind looking into them

Anonymous

Anonymous Sun 14 May 2023 05:48:20 No.7374507 Report

Quoted By: >>7374522

>>7374505
How about pija or kiliu?

Anonymous

Anonymous Sun 14 May 2023 05:51:52 No.7374511 Report

Quoted By:

>>7374500
btfo

Anonymous

Anonymous Sun 14 May 2023 05:57:17 No.7374522 Report

Quoted By: >>7374526

>>7374507
pija as in pija_(pianiishimo) on the boorus? if that's the one, i'll put it on the backburner

Anonymous

View Same Google ImgOps iqdb SauceNAO newgrid01.jpg, 1MiB, 1536x1920

Anonymous Sun 14 May 2023 05:57:24 No.7374523 Report

Quoted By: >>7374530 >>7374799 >>7374813 >>7374867

quick update to my collection (it's been a hot minute)

added styles:
Aoinhatsu, D-Floe, Dismassd, Fukurou (Owl222), Kurowa, Legendofnerd, Massakasama, Mystical, Orenji (Wholesomeorenji), Rushzilla, Shift (Waage), Tomiokasena

link here:
https://mega.nz/folder/TZElCCIL#OEutCsJPD0G5lXPXLGgfcQ

Anonymous

View Same Google ImgOps iqdb SauceNAO 1668985398172696.png, 511KiB, 516x728

Anonymous Sun 14 May 2023 05:59:11 No.7374526 Report

Quoted By:

>>7374522
Yeah that's the guy

Anonymous

View Same Google ImgOps iqdb SauceNAO tmptblscvgv.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 05:59:29 No.7374527 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 06:01:37 No.7374530 Report

Quoted By:

>>7374523
>D-Floe
Nice. Thanks for the LoRA dump as usual.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00100-640914371.png, 1MiB, 1536x1024

Anonymous Sun 14 May 2023 06:07:46 No.7374541 Report

Quoted By:

>>7374468
I will list some more features just in case anybody else needs more help. I usually just keep everything on the default except I make sure these settings are chosen
>Masked Mode: Inpaint Masked
>Masked Content: Original
>Inpaint Area: Only Masked
>Seed: Random
For the width and height I use 768 x 512 or vice versa for a portrait aspect ratio. This helps to save a ton of time compared to higher resolutions.
You don't need to always have these options selected but if you're just starting out and want to fix your eyes or fix some hands then this is a good place to start.
Later on when you want to experiment you can do as you like. The amount of space that the tool needs varies but I find that inpainting produces better results the more space you feed it. When I want to fix eyes I block off the entire face. If I want to reroll fingers i would sometimes mask off the entire hand, the wrist or even the entire arm.

btw that catbox was a missprompt and full of errors too so my bad.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00208-145095781.png, 556KiB, 696x704

Anonymous Sun 14 May 2023 06:09:20 No.7374543 Report

Quoted By:

BEHOLD ME THE MASTER OF POSES

Anonymous

View Same Google ImgOps iqdb SauceNAO almost.jpg, 114KiB, 933x1228

Anonymous Sun 14 May 2023 06:16:15 No.7374548 Report

Quoted By:

Little bit of cute yuri

Anonymous

View Same Google ImgOps iqdb SauceNAO 00846-3243120780,.jpg, 173KiB, 1024x1536

Anonymous Sun 14 May 2023 06:23:40 No.7374551 Report

Quoted By: >>7374569

Anonymous

View Same Google ImgOps iqdb SauceNAO 00766-969875766.jpg, 168KiB, 1024x1536

Anonymous Sun 14 May 2023 06:27:09 No.7374559 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 06:32:52 No.7374567 Report

Quoted By:

>spend the past hour or two dicking with powershell, cuda, visual studio since I was getting a xformers not built with cuda error
>slap on --reinstall-xformers since I didn't yet
>just works
yeah... keep it simple stupid...

Anonymous

View Same Google ImgOps iqdb SauceNAO 65d568ae7915fb5d020758dac9cfbb36.png, 658KiB, 2101x1640

Anonymous Sun 14 May 2023 06:33:51 No.7374568 Report

Quoted By: >>7374572 >>7374586 >>7374694

Now we can truly mimic real artists.

Anonymous

View Same Google ImgOps iqdb SauceNAO tmpfh8gtffx.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 06:34:46 No.7374569 Report

Quoted By: >>7374578 >>7374667

i dont even play this game but i like the character design.

>>7374551
hot. Catbox?

Anonymous

Anonymous Sun 14 May 2023 06:41:11 No.7374572 Report

Quoted By: >>7374573

>>7374568
insane, probably even more powerful with multicontrolnet

Anonymous

Anonymous Sun 14 May 2023 06:43:16 No.7374573 Report

Quoted By:

>>7374572
It's not functionally different from the inpaint trick

Anonymous

Anonymous Sun 14 May 2023 06:44:26 No.7374575 Report

Quoted By:

Is regional prompting just broken?

Anonymous

View Same Google ImgOps iqdb SauceNAO mb2.jpg, 2MiB, 3584x1344

Anonymous Sun 14 May 2023 06:48:11 No.7374576 Report

Quoted By: >>7374616 >>7374886

Anonymous

View Same Google ImgOps iqdb SauceNAO 00857-50125590 (2).jpg, 165KiB, 1024x1536

Anonymous Sun 14 May 2023 06:49:47 No.7374578 Report

Quoted By: >>7374628

>>7374569
https://files.catbox.moe/y9dfyl.jpg

Anonymous

Anonymous Sun 14 May 2023 06:51:22 No.7374580 Report

Quoted By:

>>7374292
very nice

Anonymous

View Same Google ImgOps iqdb SauceNAO 00092-3604023405.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 06:57:52 No.7374582 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 06:59:14 No.7374586 Report

Quoted By:

>>7374568
if only it worked with different models
what the reference gives is what you get

Anonymous

Anonymous Sun 14 May 2023 07:03:02 No.7374590 Report

Quoted By:

man even going a bit over 1 weight on reference makes it barf hands and tongues

Anonymous

Anonymous Sun 14 May 2023 07:05:37 No.7374596 Report

Quoted By: >>7374602

What's the hot new jap website to peddle my AI images?

Anonymous

Anonymous Sun 14 May 2023 07:06:44 No.7374598 Report

Quoted By: >>7374928

>>7374490
catbox?

Anonymous

View Same Google ImgOps iqdb SauceNAO 1684025760086277.jpg, 101KiB, 649x622

Anonymous Sun 14 May 2023 07:08:31 No.7374602 Report

Quoted By: >>7374604 >>7374612 >>7374639 >>7374758

>>7374596
None.

Anonymous

Anonymous Sun 14 May 2023 07:09:07 No.7374604 Report

Quoted By: >>7374639 >>7374642

>>7374602
what did he mean by this?

Anonymous

Anonymous Sun 14 May 2023 07:15:18 No.7374612 Report

Quoted By:

>>7374602
human rights are cool, the american social justice crowd are pure cancer though and trans is a mental illness, not an identity

Anonymous

View Same Google ImgOps iqdb SauceNAO 00076-3778423024.png, 546KiB, 896x504

Anonymous Sun 14 May 2023 07:19:38 No.7374615 Report

Quoted By:

Experimenting with all the tools is fun.

Anonymous

Anonymous Sun 14 May 2023 07:20:25 No.7374616 Report

Quoted By: >>7374621

>>7374576
which tripfag mix is this?

Anonymous

Anonymous Sun 14 May 2023 07:23:19 No.7374621 Report

Quoted By: >>7374639 >>7374886

>>7374616
Cstaber

Anonymous

Anonymous Sun 14 May 2023 07:25:43 No.7374628 Report

Quoted By: >>7374649

>>7374578
Nice gape, inpaint?

Anonymous

Anonymous Sun 14 May 2023 07:31:00 No.7374639 Report

Quoted By:

>>7374602
>>7374604
(ambassador:1.3) spreading globohomo to a culturally rich country... hmmm
fod Rahm gimme back my tax money and go to jail like Blago

>>7374621
nah the sameface is on point but the tits are too big and there's no smoke coming out of my computer

Anonymous

View Same Google ImgOps iqdb SauceNAO 1681261840714410.png, 4MiB, 1728x1728

Anonymous Sun 14 May 2023 07:33:45 No.7374642 Report

Quoted By: >>7375200

>>7374604
>human rights are cool
they're used to peddle all kinds of bullshit from illegals to twitter's censorship. they need to be more ironclad about where they begin and end.

Anonymous

Anonymous Sun 14 May 2023 07:34:08 No.7374643 Report

Quoted By:

>>7373412
cute
what did you use?

Anonymous

View Same Google ImgOps iqdb SauceNAO 2725568822.png, 975KiB, 1024x1024

Anonymous Sun 14 May 2023 07:37:16 No.7374645 Report

Quoted By:

happy mother's day /h/

Anonymous

Anonymous Sun 14 May 2023 07:37:44 No.7374646 Report

Quoted By: >>7374674 >>7374679

>>7373470
I've had good results with AdamW and batch size 3, and poor results with batch size of 8-20.

What's the consensus on a good Dadapt batch size? Dadapt is producing good results with batch size 6-7 but I haven't tried other settings.

Anonymous

Anonymous Sun 14 May 2023 07:42:29 No.7374649 Report

Quoted By: >>7374720

>>7374628
nope, all of these are pure txt2img

Anonymous

View Same Google ImgOps iqdb SauceNAO 00091-3555085664.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 07:44:57 No.7374654 Report

Quoted By: >>7374656 >>7374886 >>7375151

Baked a bra visible through clothes lora, since prompting that token is wildly inconsistent if it works at all due to how strong the 'bra' token is.

Going to rebake and separate wet clothes from non-wet clothes, since it seems to be affecting the quality of the gens.

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765641-1384267083-kill (...).png, 1MiB, 1024x1408

Anonymous Sun 14 May 2023 07:45:57 No.7374655 Report

Quoted By: >>7374672 >>7374923

good morning anons
>>7373947
>>7374183
very nice stuffs

Anonymous

Anonymous Sun 14 May 2023 07:46:35 No.7374656 Report

Quoted By:

>>7374654
Sounds great, I love that tag. Good thing you realized there's a want for non-wet see-through

Anonymous

Anonymous Sun 14 May 2023 08:00:53 No.7374667 Report

Quoted By:

>>7374569
catbox?

Anonymous

Anonymous Sun 14 May 2023 08:05:40 No.7374672 Report

Quoted By: >>7374683

>>7374655
Is there a lora for that pose? I've seen a lot of posts with it but the one on civit looks like ass

Anonymous

Anonymous Sun 14 May 2023 08:06:17 No.7374674 Report

Quoted By:

>>7374646
>d-adapt
its worse than adamW
>d-adapt batch size
theres no reason it would interact with batches differently

Anonymous

Anonymous Sun 14 May 2023 08:09:59 No.7374679 Report

Quoted By: >>7374914

>>7374646
Not reading that whole thread but is adamw the go-to now? I've always used 8bit

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765608-1328072765-high (...).png, 2MiB, 1280x1760

Anonymous Sun 14 May 2023 08:12:26 No.7374683 Report

Quoted By: >>7374687 >>7375088

>>7374672
apparently its called paizuri invitation. i dont remember where i got it but since its a small lora i can just catbox it so here you go
https://files.catbox.moe/jd7t52.safetensors

Anonymous

Anonymous Sun 14 May 2023 08:14:36 No.7374687 Report

Quoted By:

>>7374683
Judging from the size it's the civit one. Glad to see that it can produce good results though, will try it out

Anonymous

Anonymous Sun 14 May 2023 08:22:29 No.7374694 Report

Quoted By:

>>7374568
uh

Anonymous

Anonymous Sun 14 May 2023 08:24:55 No.7374695 Report

Quoted By:

>>7373535
Yeah which controlnet models are relevant

Anonymous

Anonymous Sun 14 May 2023 08:40:02 No.7374706 Report

Quoted By: >>7374730 >>7374761

Total progress: 80%| | 16/20 [36:08<09:36, 144.03s/it]

Anonymous

Anonymous Sun 14 May 2023 08:53:54 No.7374720 Report

Quoted By:

>>7374649
gib prompt pls

Anonymous

Anonymous Sun 14 May 2023 09:06:05 No.7374730 Report

Quoted By:

>>7374706
what are you doing

Anonymous

View Same Google ImgOps iqdb SauceNAO tmpy3kke9eu.png, 1MiB, 920x1376

Anonymous Sun 14 May 2023 09:16:35 No.7374737 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00136-2599895143.png, 545KiB, 896x504

Anonymous Sun 14 May 2023 09:40:04 No.7374752 Report

Quoted By: >>7374794

I will succeed with this openpose and region prompt combo, I WILL.

Anonymous

Anonymous Sun 14 May 2023 09:42:32 No.7374755 Report

Quoted By: >>7374764

is there a lora for the hentai god takeda hiromitsu? i am in need of hisato and my balls is filled to the brim

Anonymous

Anonymous Sun 14 May 2023 09:54:47 No.7374758 Report

Quoted By:

>>7374602
I'm gonna poooooooool

Anonymous

Anonymous Sun 14 May 2023 09:56:32 No.7374761 Report

Quoted By:

>>7374706
The couple of times something got stuck on my end it helped to just click on the command window and press space.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00008-1202315449.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 10:01:12 No.7374764 Report

Quoted By:

>>7374755
yes its on gayshit

Anonymous

Anonymous Sun 14 May 2023 10:28:18 No.7374783 Report

Quoted By: >>7374857

>>7374442
https://files.catbox.moe/p3dtlr.png

Anonymous

View Same Google ImgOps iqdb SauceNAO f9fd3915c1d1777eae092828aea51cf9.jpg, 710KiB, 1996x1378

Anonymous Sun 14 May 2023 10:35:55 No.7374794 Report

Quoted By: >>7374811 >>7374981

>>7374752
God damn it I forgot how to do latent upscaling again..

Anonymous

Anonymous Sun 14 May 2023 10:40:26 No.7374799 Report

Quoted By:

>>7374523
cool

Anonymous

Anonymous Sun 14 May 2023 10:46:30 No.7374811 Report

Quoted By:

>>7374794
Upping denoising to 0.5 minimum is a good start

Anonymous

Anonymous Sun 14 May 2023 10:49:38 No.7374813 Report

Quoted By: >>7375055

>>7374523
We need someone to take over gayshit

Anonymous

Anonymous Sun 14 May 2023 11:11:15 No.7374843 Report

Quoted By: >>7374848

lol I found a Gradio bug that can reproducibly force kernel error on the host machine.

Anonymous

Anonymous Sun 14 May 2023 11:14:18 No.7374848 Report

Quoted By:

>>7374843
Gradio HATE

Anonymous

Anonymous Sun 14 May 2023 11:32:17 No.7374857 Report

Quoted By:

>>7374783
Cool, thank you

Anonymous

Anonymous Sun 14 May 2023 12:03:55 No.7374867 Report

Quoted By:

>>7374523
based

Anonymous

Anonymous Sun 14 May 2023 12:07:54 No.7374869 Report

Quoted By: >>7374883 >>7374886

hello anons
i was wondering if there is a way to check the tags of stable diffusion images
I generated a lot of pictures and wish to see the tags of those pictures is there some module or something to check the tags?

thank you anons.

Anonymous

View Same Google ImgOps iqdb SauceNAO 10670-2197860493-face, gawr gura (...).png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 12:22:08 No.7374880 Report

Quoted By: >>7374885

>>7374303
>>7374313
nothing wrong with NAI if you use style loras
picrel

>>7374474
something that touched upon : what if variation seed?

Anonymous

Anonymous Sun 14 May 2023 12:24:46 No.7374883 Report

Quoted By: >>7374931

>>7374869
PNG Info tab

Anonymous

Anonymous Sun 14 May 2023 12:24:55 No.7374884 Report

Quoted By:

bros help this shits conditioning me like a pavlovian coomer

Anonymous

Anonymous Sun 14 May 2023 12:25:03 No.7374885 Report

Quoted By: >>7374892

>>7374880
no way this was made with nai

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_ibzkg3.png, 3MiB, 1024x1536

Anonymous Sun 14 May 2023 12:25:34 No.7374886 Report

Quoted By: >>7374890 >>7374931

>>7374576
catbox for the left two?

>>7374621
The last (and only) time I used cstabber I got naked toddlers and immediately deleted it and the gens it made

>>7374654
TYFYS

>>7374869
PNG Info tab at the top

Anonymous

Anonymous Sun 14 May 2023 12:31:01 No.7374890 Report

Quoted By:

>>7374886
Just put lolicon in the negative prompt anon

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_kk9lzq.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 12:31:21 No.7374892 Report

Quoted By: >>7374896 >>7374903

>>7374885
checked, and I indeed posted from the wrong batch since I didn't have webui open to check
here's one from NAI with the same style loras

Anonymous

Anonymous Sun 14 May 2023 12:33:01 No.7374896 Report

Quoted By: >>7374903

>>7374892
Catbox link doesn't seem to work

Anonymous

View Same Google ImgOps iqdb SauceNAO 00111-3098623904.png, 1MiB, 1088x1392

Anonymous Sun 14 May 2023 12:35:22 No.7374900 Report

Quoted By:

I didn't specify a prompt for multiple viewpoints, but I made a good work

Anonymous

View Same Google ImgOps iqdb SauceNAO 01079-mistoonAnime_v10_2449443821.png, 406KiB, 512x512

Anonymous Sun 14 May 2023 12:35:50 No.7374901 Report

Quoted By:

you gotta have your protein if you wanna be a big strong boy

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_ev2wzk.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 12:36:19 No.7374903 Report

Quoted By: >>7374909 >>7374996

>>7374892
another one, other style loras

>>7374896
skill issue, works on my machine
here's the copypaste : https://rentry.org/6cdpw

Anonymous

Anonymous Sun 14 May 2023 12:36:51 No.7374904 Report

Quoted By:

Anyone know of a lora for hentai pussies?
I'm only getting axewounds with the inpainting.

Anonymous

Anonymous Sun 14 May 2023 12:40:10 No.7374909 Report

Quoted By: >>7374913

>>7374903
>skill issue
Yeah it was, thanks.
>DDetailer
Is this worth using?

Anonymous

Anonymous Sun 14 May 2023 12:42:13 No.7374911 Report

Quoted By:

I know I'm going to get rocks thrown at me for asking this and I totally deserve it but, how do you actually setup username and password for the StableDiffusionUI_Voldermort_paperspace edition?

No matter how much I try setting the gradio_auth etc, I can't get it to work.

Anonymous

View Same Google ImgOps iqdb SauceNAO 1656626009594439.png, 1MiB, 752x1136

Anonymous Sun 14 May 2023 12:43:56 No.7374912 Report

Quoted By:

Trying to make a fei lora, but it's still pretty wonky more often than not. Trying to stick to colorized images with a couple cropped doujin pages, but it's the smallest dataset I've worked with(92), so it probably needs some training adjustments. Not really helping that a lot of his colored work has kind of a natural burn to it and maybe I need to be more picky

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_plfkwg.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 12:44:38 No.7374913 Report

Quoted By: >>7374919

>>7374909
>DDDetailer
Yeah, it's very useful for fixing faces, takes less time than inapainting and doesn't really leave seams.
The only problem is that it can't do batches, so you gotta make them one by one. I'm a VRAMlet so it doesn't bother me too much but still.

Anonymous

Anonymous Sun 14 May 2023 12:45:35 No.7374914 Report

Quoted By: >>7374945

>>7374679
8bit is a vram optimization
You were probably using 8bit AdamW
I have lots of vram so just turned 8bit off

Anonymous

Anonymous Sun 14 May 2023 12:52:22 No.7374919 Report

Quoted By:

>>7374913
Damn...

Anonymous

Anonymous Sun 14 May 2023 12:57:16 No.7374923 Report

Quoted By: >>7375088

>>7374655
CATBOOOOOOOOOOOOOOOOX pls

Anonymous

Anonymous Sun 14 May 2023 13:00:41 No.7374928 Report

Quoted By:

>>7374598
ah this is just a controlnet thing , nothing unique about the prompt

Anonymous

View Same Google ImgOps iqdb SauceNAO 00018-497734726.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 13:02:03 No.7374930 Report

Quoted By:

>>7374424
yeah

Anonymous

Anonymous Sun 14 May 2023 13:04:22 No.7374931 Report

Quoted By: >>7374941

>>7374883
>>7374886
thanks anons do any of you know any cloud stable diffusion services
my 3050 is dying

Anonymous

Anonymous Sun 14 May 2023 13:05:11 No.7374932 Report

Quoted By: >>7374965

has anyone been able to create a unique style that looks good by mixing artist loras?

Anonymous

View Same Google ImgOps iqdb SauceNAO 12586624452.jpg, 1MiB, 3436x2457

Anonymous Sun 14 May 2023 13:05:31 No.7374934 Report

Quoted By: >>7374968

Anonymous

Anonymous Sun 14 May 2023 13:07:53 No.7374936 Report

Quoted By:

you know what i think someone should do is see if they can form some sort of symbiosis with the AI.
>look at picture try to copy it by hand
>make a whole lot of these
>train controlnet so so that it outputs pictures from your sketches

would something like this work?

Anonymous

Anonymous Sun 14 May 2023 13:09:15 No.7374941 Report

Quoted By:

>>7374931
irc the waifudiffusion guys used coreweave
thats the only one i know

Anonymous

Anonymous Sun 14 May 2023 13:11:26 No.7374943 Report

Quoted By:

now that i think about it, can you use multiple models with controlnet?
like having both the pose and scribble?

Anonymous

Anonymous Sun 14 May 2023 13:13:47 No.7374945 Report

Quoted By: >>7375289 >>7375296

>>7374914
the bait is getting more harmful by the day itt.
non 8bit gives horrid results

Anonymous

Anonymous Sun 14 May 2023 13:39:11 No.7374965 Report

Quoted By:

>>7374932
I imagine that depends on how you define unique and how pedantic you want to be when it comes to certain styles resembling other styles

Anonymous

Anonymous Sun 14 May 2023 13:41:30 No.7374968 Report

Quoted By: >>7374994

>>7374934
Your params, model and loras, now

Anonymous

Anonymous Sun 14 May 2023 13:50:19 No.7374972 Report

Quoted By: >>7374975 >>7374976

Is it typically better to prioritise weight for character loras over poses or vice versa?

Anonymous

Anonymous Sun 14 May 2023 13:52:23 No.7374975 Report

Quoted By:

>>7374972
Depends heavily on the loras
Just play around with the values

Anonymous

Anonymous Sun 14 May 2023 13:55:37 No.7374976 Report

Quoted By:

>>7374972
Depends. I train my stuff to work at 1.0, but I'd lower style first on others.

Anonymous

View Same Google ImgOps iqdb SauceNAO sgsd.png, 2MiB, 2048x1152

Anonymous Sun 14 May 2023 13:59:38 No.7374981 Report

Quoted By: >>7374983 >>7374988

>>7374794
I give up, upscaling just simply doesn't work for the soft painterly look I'm after.

Anonymous

Anonymous Sun 14 May 2023 14:00:40 No.7374983 Report

Quoted By: >>7374987

>>7374981
Isn't that because you have no prompt? What did you expect.

Anonymous

Anonymous Sun 14 May 2023 14:02:49 No.7374987 Report

Quoted By: >>7374991

>>7374983
Even if I add a prompt, change the denoise, it gets fucked by the lack of detail. It's like streaming at a low bitrate and the image has too much detail in it, blurry mess.

Anonymous

Anonymous Sun 14 May 2023 14:05:00 No.7374988 Report

Quoted By: >>7375026

>>7374981
Is this the base image you're posting? Or is this post-upscale?

Anonymous

Anonymous Sun 14 May 2023 14:05:59 No.7374991 Report

Quoted By: >>7375026

>>7374987
Oh yeah, I get what you mean.
Upscaling blurry parts gets you those weird sharp cutoffs, even on regular img2img.

Anonymous

View Same Google ImgOps iqdb SauceNAO 353029333006211.png, 392KiB, 470x685

Anonymous Sun 14 May 2023 14:07:02 No.7374992 Report

Quoted By: >>7374997 >>7375010 >>7375028

If I were to train a LORA on a character's VN sprites, would it wind up overfit in a way that mattered? I DO want them in similar poses/the same poses, but different outfits/lack of outfits. I'm just kind of at my wit's end trying to figure out a good solution.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00118-2461969560-(extremely deta (...).png, 898KiB, 768x1160

Anonymous Sun 14 May 2023 14:08:33 No.7374993 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 14:09:24 No.7374994 Report

Quoted By:

>>7374968
https://files.catbox.moe/4k3nlj.png
That's the original before upscaling twice and inpainting the eyes

Anonymous

Anonymous Sun 14 May 2023 14:09:47 No.7374996 Report

Quoted By: >>7375002 >>7375029

>>7374903
HNNNGH, super cute. Glad there's more chubbyfags here.

Anonymous

Anonymous Sun 14 May 2023 14:10:11 No.7374997 Report

Quoted By: >>7375028

>>7374992
Not related, but remember to convert to jpg because training treats transparent backgrounds as pure black, and having lots of images with pitch black backgrounds in them affect loras negatively.

Might have changed now, but that was the word on the street last I checked.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_pkl69w.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 14:13:13 No.7375002 Report

Quoted By:

>>7374996
I've been posting these on /vt/

Anonymous

View Same Google ImgOps iqdb SauceNAO 00645-2433898932-1girl, hair orn (...).png, 610KiB, 640x896

Anonymous Sun 14 May 2023 14:21:21 No.7375010 Report

Quoted By:

>>7374992
cutetsuki!

Anonymous

Anonymous Sun 14 May 2023 14:30:31 No.7375022 Report

Quoted By: >>7375023

Been getting this error whenever I try to edit a Lora, metadata, anyone got an idea on how to fix it?

Anonymous

View Same Google ImgOps iqdb SauceNAO Error.png, 16KiB, 1474x113

Anonymous Sun 14 May 2023 14:31:02 No.7375023 Report

Quoted By:

>>7375022
Im retarded.

Anonymous

Anonymous Sun 14 May 2023 14:33:13 No.7375026 Report

Quoted By:

>>7374988
Post upscale via photoshop.

>>7374991
Guess I'll abandon the style..

Anonymous

Anonymous Sun 14 May 2023 14:37:03 No.7375028 Report

Quoted By:

>>7374997
Not sure how much this matters, I dont think I noticed much of a difference and I've been training on visual novel contents pretty much exclusively.

>>7374992
>at my wit's end
whats the issue?
You probably should not include EVERY sprite variation, just pick a few expressions from each pose. There should be no issues switching clothes or poses if you tagged things properly and didn't use some weird settings

Anonymous

Anonymous Sun 14 May 2023 14:38:09 No.7375029 Report

Quoted By: >>7375037

>>7374996
>chubby
im into chubby but thats BBW area, maybe im just not horny but my immediate thought is that its fucking disgusting

Anonymous

Anonymous Sun 14 May 2023 14:48:30 No.7375037 Report

Quoted By: >>7375411

>>7375029
I cure myself of chubby hornymind every time by thinking about people IRL with that body type.

Anonymous

Anonymous Sun 14 May 2023 14:58:02 No.7375050 Report

Quoted By: >>7375052 >>7375085 >>7375104 >>7375105 >>7375458

>>7373269
Can somebody explain why does the recent sd version updated to torch 2.0 eats up much more vram? Previously my shitty old rtx 2080 could handle generating pics up to 1024x1280 but now it tops at 960p, anything more than that gives me an out of memory error. It takes much more time to generate too, sometimes getting stuck for 15+ seconds on 100%

Anonymous

Anonymous Sun 14 May 2023 14:59:21 No.7375052 Report

Quoted By: >>7375071

>>7375050
do a clean install

Anonymous

Anonymous Sun 14 May 2023 15:00:28 No.7375055 Report

Quoted By: >>7375080

>>7374813
t. webui fork user

Anonymous

View Same Google ImgOps iqdb SauceNAO 00013-2925714946.png, 1MiB, 768x1152

Anonymous Sun 14 May 2023 15:01:48 No.7375058 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 15:15:24 No.7375071 Report

Quoted By:

>>7375052
I already deleted and set up a new venv and reinstalled all dependencies, what would a clean install change?

Anonymous

Anonymous Sun 14 May 2023 15:17:47 No.7375075 Report

Quoted By: >>7375081

Today I will learn there is a difference between nude and totally nude.

Anonymous

Anonymous Sun 14 May 2023 15:17:56 No.7375076 Report

Quoted By: >>7375084 >>7375096

Have things improved since last year? Can I generate fappable kitsunes and feet now?

Anonymous

Anonymous Sun 14 May 2023 15:20:06 No.7375080 Report

Quoted By:

>>7375055
I'm actually on april commit but it just werks

Anonymous

Anonymous Sun 14 May 2023 15:20:39 No.7375081 Report

Quoted By:

>>7375075
completely nude*

Anonymous

Anonymous Sun 14 May 2023 15:21:40 No.7375084 Report

Quoted By:

>>7375076
kek

Anonymous

Anonymous Sun 14 May 2023 15:21:41 No.7375085 Report

Quoted By: >>7375099 >>7375105 >>7375458

>>7375050
Vram usage increased for me too, though at the time I blamed it on switching to sdp from xformers. Perhaps it was the torch update instead.

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765619-2591400437-kill (...).png, 1MiB, 1280x1760

Anonymous Sun 14 May 2023 15:24:45 No.7375088 Report

Quoted By: >>7375434

>>7374923
sure heres original one
https://files.catbox.moe/qh4qvp.png
and heres edited one
https://files.catbox.moe/fwn8hl.png
splashbrush and killjoy lora you can find it in gitgud, i made ginseng lora but its just ok and paizuri inv is in this >>7374683

Anonymous

Anonymous Sun 14 May 2023 15:25:50 No.7375091 Report

Quoted By: >>7375096 >>7375100 >>7375106

Does generating ai pics damage the graphics card?

Anonymous

Anonymous Sun 14 May 2023 15:28:34 No.7375096 Report

Quoted By: >>7375123

>>7375076
>>7375091
go back to sleep natard

Anonymous

Anonymous Sun 14 May 2023 15:29:58 No.7375099 Report

Quoted By: >>7375105

>>7375085
I didn't get any differences between sdp and xformers, and the different torch versions.
I wouldn't be surprised if Voldy fucked something up again.

Anonymous

Anonymous Sun 14 May 2023 15:32:21 No.7375100 Report

Quoted By:

>>7375091
Yes, I have the full install on a USB drive that I always bring with me to the public library whenever I want to proompt

Anonymous

Anonymous Sun 14 May 2023 15:34:47 No.7375104 Report

Quoted By: >>7375545

>>7375050
Did you change upscaler?

For me, I ran out of VRAM on 4xAnimeSharp but can do same res on Latent(Nearest).

Some upscalers seem to eat more vram.

Anonymous

Anonymous Sun 14 May 2023 15:35:17 No.7375105 Report

Quoted By: >>7375546

>>7375085
>>7375099
Well, I just tried adding --medvram and now it's similar to like it was before the update, maybe it can handle even higher res. Also it doesn't even seem to affect generation time pretty much at all, unlike when I was using medvram back in September (because I had no clue what I was doing back then)
t. >>7375050

Anonymous

Anonymous Sun 14 May 2023 15:36:04 No.7375106 Report

Quoted By: >>7375114

>>7375091
Yes, it causes coil whine that vibrates smd's off the pcb. To prevent the card from self destructing, you need to put the card in an oven at 500F or more to remelt the solder that holds the components on the pcb after every 24 hours of use.

Anonymous

Anonymous Sun 14 May 2023 15:40:15 No.7375114 Report

Quoted By:

>>7375106
Instructions unclear, put your dog in the oven. Do you want the meat or nah?

Anonymous

Anonymous Sun 14 May 2023 15:41:16 No.7375116 Report

Quoted By: >>7375119 >>7375120 >>7375123 >>7375133

Is there a guide to generate good lolis?

Anonymous

View Same Google ImgOps iqdb SauceNAO pedo file.png, 13KiB, 256x256

Anonymous Sun 14 May 2023 15:42:44 No.7375119 Report

Quoted By:

>>7375116
Good morning sirs!

Anonymous

Anonymous Sun 14 May 2023 15:42:50 No.7375120 Report

Quoted By: >>7375152

>>7375116
https://civitai.com/models/40016/glowing-art
This lora

Anonymous

Anonymous Sun 14 May 2023 15:43:36 No.7375123 Report

Quoted By: >>7375125

>>7375116
I told you already.
>>7375096

Anonymous

Anonymous Sun 14 May 2023 15:44:23 No.7375125 Report

Quoted By: >>7375132

>>7375123
Shut up, faggot.

Anonymous

Anonymous Sun 14 May 2023 15:46:02 No.7375128 Report

Quoted By: >>7375129 >>7375130

what was the program you guys use to filter out similar images?

Anonymous

Anonymous Sun 14 May 2023 15:46:18 No.7375129 Report

Quoted By: >>7375280

>>7375128
Czkawka.

Anonymous

Anonymous Sun 14 May 2023 15:47:18 No.7375130 Report

Quoted By:

>>7375128
dupeguru, irfanview

Anonymous

Anonymous Sun 14 May 2023 15:47:43 No.7375131 Report

Quoted By: >>7375135 >>7375138

Off topic but is there something like stablediffusion but for fanfics, without the "no erotic" faggotry of chatgpt?

Anonymous

Anonymous Sun 14 May 2023 15:47:46 No.7375132 Report

Quoted By: >>7375134

>>7375125
nah
now go back

Anonymous

Anonymous Sun 14 May 2023 15:48:25 No.7375133 Report

Quoted By:

>>7375116
Say the tripfags name three times in the thread to summon some glowing mixes

Anonymous

Anonymous Sun 14 May 2023 15:48:43 No.7375134 Report

Quoted By:

>>7375132
nah

Anonymous

Anonymous Sun 14 May 2023 15:48:51 No.7375135 Report

Quoted By:

>>7375131
>is there
Yes.
>equal to any proprietary solution other than nai
No.

Anonymous

Anonymous Sun 14 May 2023 15:51:36 No.7375138 Report

Quoted By:

>>7375131
/g/lmg/

Anonymous

View Same Google ImgOps iqdb SauceNAO 00082-3900413254.png, 953KiB, 768x1152

Anonymous Sun 14 May 2023 15:55:15 No.7375143 Report

Quoted By:

Did mixer anon ever post/release his mixes anywhere? I've been trying to get a model that does anime but is capable of having those detailed skin textures. No luck.

Pic unrelated.

Anonymous

Anonymous Sun 14 May 2023 15:55:33 No.7375144 Report

Quoted By: >>7375145 >>7375149 >>7375150 >>7375161 >>7375172 >>7375201 >>7375205

https://civitai.com/models/65990/queens-stallion
w-what

Anonymous

Anonymous Sun 14 May 2023 15:55:59 No.7375145 Report

Quoted By: >>7375148

>>7375144
one of /d/'s worst fetish

Anonymous

Anonymous Sun 14 May 2023 15:56:46 No.7375148 Report

Quoted By:

>>7375145
I genuinely cannot comprehend how anyone finds that hot

Anonymous

Anonymous Sun 14 May 2023 15:57:30 No.7375149 Report

Quoted By:

>>7375144
based

Anonymous

Anonymous Sun 14 May 2023 15:58:29 No.7375150 Report

Quoted By:

>>7375144
Did this come from the pingu blowjob face? Who comes up with this stuff?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00132-342284176.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 15:58:33 No.7375151 Report

Quoted By: >>7375156 >>7375416

>>7374654
Rebaked bra visible through clothes for the dry version. I'll redo the wet shirts another time with a dataset for that specific purpose.

>https://mega.nz/folder/ZvA21I7L#ZZzU42rdAyWFOWQ_O94JaQ/folder/0io1WQTa
Get your see-through shirts here.

Anonymous

Anonymous Sun 14 May 2023 15:59:46 No.7375152 Report

Quoted By:

>>7375120
Why is it all glowing? What's the point?

Anonymous

Anonymous Sun 14 May 2023 15:59:47 No.7375153 Report

Quoted By: >>7375154

How do I use lora training?
After following the options in train_XTI.bat it simply closes and nothing happens

Anonymous

Anonymous Sun 14 May 2023 16:00:53 No.7375154 Report

Quoted By:

>>7375153
go. back. to. sleep.

Anonymous

Anonymous Sun 14 May 2023 16:02:05 No.7375156 Report

Quoted By:

>>7375151
Doing gods work, anon

Anonymous

Anonymous Sun 14 May 2023 16:03:08 No.7375157 Report

Quoted By: >>7375160

I scrapped my lora folder in a moment of post nut clarity and now I'm having to download them all again ;-;

Anonymous

Anonymous Sun 14 May 2023 16:04:28 No.7375160 Report

Quoted By: >>7375168

>>7375157
It really do be like that...
What was it?

Anonymous

Anonymous Sun 14 May 2023 16:04:53 No.7375161 Report

Quoted By: >>7375172

>>7375144
Holy shit, that's actually kinda sexy.
Certainly better than the fatso obsessed, at least.

Anonymous

Anonymous Sun 14 May 2023 16:08:15 No.7375168 Report

Quoted By: >>7375174

>>7375160
The realisation I'd sat for like two hours fine tuning an AI to get off hit me like a tonne of fucking bricks for some reason

Anonymous

View Same Google ImgOps iqdb SauceNAO 00010-3542049917.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 16:08:34 No.7375169 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 16:09:02 No.7375172 Report

Quoted By: >>7375173

>>7375144
>>7375161
Society.

Anonymous

Anonymous Sun 14 May 2023 16:09:34 No.7375173 Report

Quoted By:

>>7375172
So true

Anonymous

Anonymous Sun 14 May 2023 16:10:47 No.7375174 Report

Quoted By: >>7375178

>>7375168
Get off to what?

Anonymous

Anonymous Sun 14 May 2023 16:12:20 No.7375178 Report

Quoted By: >>7375185

>>7375174
I purged my output folder with it, it was a bunch of hypnosis shit

Anonymous

View Same Google ImgOps iqdb SauceNAO 00013-558344237.png, 692KiB, 768x1024

Anonymous Sun 14 May 2023 16:14:14 No.7375183 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 16:16:32 No.7375185 Report

Quoted By:

>>7375178
>hypnosis
Kek.

Anonymous

View Same Google ImgOps iqdb SauceNAO nyaruko.png, 92KiB, 600x337

Anonymous Sun 14 May 2023 16:24:21 No.7375191 Report

Quoted By: >>7375193 >>7375202

If I install this will I be able to create Nyaruko porn right off the bat?
Or do I have to train it?

Anonymous

Anonymous Sun 14 May 2023 16:25:51 No.7375193 Report

Quoted By: >>7375198

>>7375191
https://civitai.com/models/65344/haiyore-nyaruko-san

Anonymous

Anonymous Sun 14 May 2023 16:30:52 No.7375198 Report

Quoted By:

>>7375193
!!!

Anonymous

Anonymous Sun 14 May 2023 16:36:59 No.7375200 Report

Quoted By:

>>7374642
I thought it was obvious - do whatever you want so long as it isn't harming another person. Wasn't that what we all wanted? I thought the problem was that the right wants to control everyone. Is that no longer true?

Anonymous

Anonymous Sun 14 May 2023 16:38:56 No.7375201 Report

Quoted By:

>>7375144
does this work with character loras? could do some nice trolling

Anonymous

Anonymous Sun 14 May 2023 16:43:38 No.7375202 Report

Quoted By:

>>7375191
Skies the limit anon, just grab based 64, the novelai VAE and do your initial config and with the power of loras + example prompts you now have unlimited coomer fuel

Anonymous

Anonymous Sun 14 May 2023 16:48:45 No.7375205 Report

Quoted By:

>>7375144
in tarnation

Anonymous

View Same Google ImgOps iqdb SauceNAO 00675-775354-1girl, long hair, h (...).png, 611KiB, 640x896

Anonymous Sun 14 May 2023 16:51:48 No.7375207 Report

Quoted By: >>7375330

Testing Crimsoncomics.
It's gonna be kino.

Anonymous

Anonymous Sun 14 May 2023 17:00:48 No.7375216 Report

Quoted By: >>7375225

Are traps allowed in this thread?

Anonymous

Anonymous Sun 14 May 2023 17:03:11 No.7375218 Report

Quoted By: >>7375227

I spent hours making the perfect lora mix, but now I realize it can't generate males. Not even a pov crotch, only floating dicks.

Anonymous

Anonymous Sun 14 May 2023 17:04:55 No.7375225 Report

Quoted By:

>>7375216
Yes, once.

Anonymous

Anonymous Sun 14 May 2023 17:06:56 No.7375227 Report

Quoted By:

>>7375218
best bet is to go to /d/

Anonymous

View Same Google ImgOps iqdb SauceNAO kotoboobie.jpg, 221KiB, 1024x1536

Anonymous Sun 14 May 2023 17:13:41 No.7375237 Report

Quoted By: >>7375261 >>7375421

Anonymous

Anonymous Sun 14 May 2023 17:36:37 No.7375261 Report

Quoted By: >>7375421

>>7375237
Very nice. Did you draw the xray silhouette? Underused effect.

Anonymous

Anonymous Sun 14 May 2023 17:42:42 No.7375269 Report

Quoted By: >>7375274 >>7375277

I've randomly started getting the error NansException: A tensor with all NaNs was produced in VAE. This could be because there's not enough precision to represent the picture. Try adding --no-half-vae commandline argument to fix this. Use --disable-nan-check commandline argument to disable this check.

when I wasn't before, anyone run into this?

Anonymous

Anonymous Sun 14 May 2023 17:49:47 No.7375274 Report

Quoted By:

>>7375269
Based retard.

Anonymous

Anonymous Sun 14 May 2023 17:51:52 No.7375277 Report

Quoted By: >>7375292

>>7375269
yea and idk how to fix it other than changing vaes

Anonymous

Anonymous Sun 14 May 2023 17:53:38 No.7375280 Report

Quoted By: >>7375287

>>7375129
what settings works best on this?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00214-340690403.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 17:59:12 No.7375286 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 17:59:23 No.7375287 Report

Quoted By: >>7375311

>>7375280
Depends. I usually do half at 16 hash. It deletes some variations too, but I want that.

Anonymous

Anonymous Sun 14 May 2023 18:00:37 No.7375289 Report

Quoted By: >>7375294 >>7375384 >>7375391

>>7374945
Seriously? You think 8bit adamW gives better results? I've just never heard that opinion before. I always assumed it was a vramlet cope feature.

Anonymous

Anonymous Sun 14 May 2023 18:01:28 No.7375292 Report

Quoted By:

>>7375277
Thing is I never got it until earlier today, despite using the same VAE this whole time

Anonymous

View Same Google ImgOps iqdb SauceNAO 00216-332387017.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 18:02:43 No.7375294 Report

Quoted By: >>7375332

>>7375289
NTA but on my 1024x1024 LoRAs, I get better results given the same steps on 8bit adam vs dadapt.

Dadapt can give you good results, but IMO it needs way more steps and it's also a good chunk slower.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00342-datetime.png, 3MiB, 1280x1792

Anonymous Sun 14 May 2023 18:03:36 No.7375295 Report

Quoted By: >>7375484

One more post for mothers day. Btw is there a fix/known cause for that annoying hang up that sometimes happens after the last step? It's basically done, but just stops after reaching 30/30 steps and it might take minutes to continue. Only seems to happen in img2img for me.

Anonymous

Anonymous Sun 14 May 2023 18:03:51 No.7375296 Report

Quoted By: >>7375389

>>7374945
BTW, seperate topic: I also use bf16 precision. I heard it's better than fp16 but I didn't notice a difference.

Anonymous

Anonymous Sun 14 May 2023 18:04:03 No.7375298 Report

Quoted By:

Completely bored of prompting again. Time to bake

Anonymous

Anonymous Sun 14 May 2023 18:11:01 No.7375304 Report

Quoted By:

in which folder do I save my Loras?

Anonymous

Anonymous Sun 14 May 2023 18:12:45 No.7375307 Report

Quoted By:

How do you do inpainting without leaving darkened areas with mismatched colors at the borders? It looks like a water stain in the image.

Anonymous

Anonymous Sun 14 May 2023 18:15:31 No.7375311 Report

Quoted By: >>7375319

>>7375287
if you're only doing half at 16 hash you are definitely leaving duplicates in your dataset

Anonymous

Anonymous Sun 14 May 2023 18:19:09 No.7375319 Report

Quoted By: >>7375327

>>7375311
Some, sure.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00277-180606459-nsfw, scan, (tra (...).png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 18:19:43 No.7375320 Report

Quoted By: >>7375338

Anonymous

Anonymous Sun 14 May 2023 18:27:43 No.7375327 Report

Quoted By:

>>7375319
if you don't care about variations and just want to completely nuke every duplicate, 16 hash at the lowest similarity will get all of them without any false positives (besides variations). though i do like to check the matches and delete them on an individual bases. 8 at high similarity is also pretty good afterwards for double checking for duplicates drawn in different mediums (like a photo of the drawing vs the drawn-over digital version). sometimes lanczos3 and catmullrom miss different things, so it's always nice to run it through both

Anonymous

View Same Google ImgOps iqdb SauceNAO 00789-512766-1girl, short hair, (...).png, 568KiB, 640x896

Anonymous Sun 14 May 2023 18:30:51 No.7375330 Report

Quoted By: >>7375333

>>7375207
Still.

Anonymous

Anonymous Sun 14 May 2023 18:32:54 No.7375332 Report

Quoted By: >>7375334 >>7375336

>>7375294
At least you don't have to mess with fucking learning rate. dadapt is just easier for styles

Anonymous

Anonymous Sun 14 May 2023 18:33:38 No.7375333 Report

Quoted By: >>7375335

>>7375330
Do you like this picture yourself anon? Do you find it attractive? Arousing?

Anonymous

Anonymous Sun 14 May 2023 18:33:56 No.7375334 Report

Quoted By:

>>7375332
i've never understood this, it took me a day of testing to find a decent learning rate for my settings and i've never had to change it

Anonymous

Anonymous Sun 14 May 2023 18:36:09 No.7375335 Report

Quoted By:

So what was the proccess for coloring mangas with controlnet again?
>>7375333
Mhm. And nostalgic.
It's pretty neat for a random seed for a test.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00244-2901755547.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 18:38:17 No.7375336 Report

Quoted By: >>7375345

>>7375332
Oh prob, I just train charactes and concept LoRAs, haven't tried with styles.

Since I've used a lot of time testing unet and text lrs, now I can (kinda) set them easily depending of the character and the concept. Not sure how it would go for styles though.

If I knew nothing, yeah I would start with dadapt, you just have to be careful with repeats and alpha.

Anonymous

Anonymous Sun 14 May 2023 18:40:01 No.7375338 Report

Quoted By: >>7375354

>>7375320
Nice. Got catbox?

Anonymous

Anonymous Sun 14 May 2023 18:45:32 No.7375345 Report

Quoted By: >>7375359

>>7375336
yeah dadapt sucks for characters

Anonymous

Anonymous Sun 14 May 2023 18:45:33 No.7375346 Report

Quoted By: >>7375352 >>7375353 >>7375357

im over here from /b/ diffusion thread, this thread is so much nicer, what other boards have diffusion threads?

Anonymous

Anonymous Sun 14 May 2023 18:46:51 No.7375352 Report

Quoted By:

>>7375346
I think most of the red boards have one

Anonymous

Anonymous Sun 14 May 2023 18:47:15 No.7375353 Report

Quoted By:

>>7375346
>retarded op removed the links to the other boards
lol

Anonymous

Anonymous Sun 14 May 2023 18:47:23 No.7375354 Report

Quoted By: >>7375360

>>7375338
https://files.catbox.moe/xxvfzt.png

Anonymous

View Same Google ImgOps iqdb SauceNAO 00130-3229187813-.png, 1MiB, 920x1392

Anonymous Sun 14 May 2023 18:47:35 No.7375356 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 18:47:47 No.7375357 Report

Quoted By: >>7375363

>>7375346
/vtai/ is surprisingly nice if you like teasy ecchi stuff. I have never watched a single vtuber in my life and never going to but the anime girls look nice and I prompting them can be fun
/g/ has sdg which is complete trash
never really looked into other generals

Anonymous

Anonymous Sun 14 May 2023 18:47:54 No.7375359 Report

Quoted By:

>>7375345
nah

Anonymous

Anonymous Sun 14 May 2023 18:48:33 No.7375360 Report

Quoted By:

>>7375354
Thanks

Anonymous

Anonymous Sun 14 May 2023 18:49:52 No.7375361 Report

Quoted By:

is there a sinkai lora?

Anonymous

Anonymous Sun 14 May 2023 18:52:34 No.7375363 Report

Quoted By: >>7375369 >>7375378 >>7375381 >>7375395

>>7375357
I never posted on /g/ but I remember that it used to have a bunch of really cool gens with creative ideas and lots of technical prompts. The gens posted there had a lot of rich imagination behind them and they were very inspiring. I used to go there for ideas and snag technical prompts. That wasn't even more than a few months ago. Now it's filled with 1girl, x, y, z.
what happened?

Anonymous

Anonymous Sun 14 May 2023 18:55:15 No.7375369 Report

Quoted By:

>>7375363
>Now it's filled with 1girl, x, y, z.
BASED

Anonymous

View Same Google ImgOps iqdb SauceNAO 00861-646427-1girl, two side up, (...).png, 705KiB, 640x896

Anonymous Sun 14 May 2023 18:57:15 No.7375375 Report

Quoted By:

>one epoch puts in molestation not seen in the others
heh

Anonymous

Anonymous Sun 14 May 2023 19:01:43 No.7375378 Report

Quoted By:

>>7375363
Probably burnout or people feeling they had reached the limits of what the AI could do. I know the latter applied to me at some point, although this was back when dreambooths were the latest new toy

Anonymous

Anonymous Sun 14 May 2023 19:02:41 No.7375381 Report

Quoted By:

>>7375363
>what happened?
It turned into a discord chat

Anonymous

Anonymous Sun 14 May 2023 19:04:11 No.7375384 Report

Quoted By:

>>7375289
8bitadam is for nvidia iirc, the other one is for AMD so its not about vram lol

Anonymous

Anonymous Sun 14 May 2023 19:06:05 No.7375389 Report

Quoted By:

>>7375296
>difference
it doesnt really matter, iirc doing full bf16 might cuck vramlets from using ur loras though or something

Anonymous

Anonymous Sun 14 May 2023 19:08:07 No.7375391 Report

Quoted By:

>>7375289
I use 8bit adamw and I have a 4090

Anonymous

Anonymous Sun 14 May 2023 19:09:35 No.7375395 Report

Quoted By: >>7375404 >>7375412 >>7375415 >>7375463 >>7375487

>>7375363
the population of /sdg/ has been entirely replaced at this point with newer people who started with overtuned models like AOM so the need for experimenting with prompting was never there. that combined with the fact that loras take less effort and achieve way more predictable results means why would you bother with style prompts when you can just slap on a few loras and use AOM. it's just a microcosm of new technology getting easier to use and thus old knowledge being obsolete and lost

Anonymous

Anonymous Sun 14 May 2023 19:14:09 No.7375404 Report

Quoted By: >>7375413

>>7375395
newfag here, blame the guides for that
voldy guide still says to use AOM and similar even though so many are better

Anonymous

Anonymous Sun 14 May 2023 19:18:58 No.7375411 Report

Quoted By:

>>7375037
imagine the smell

Anonymous

Anonymous Sun 14 May 2023 19:19:36 No.7375412 Report

Quoted By:

>>7375395
A reasonable enough explanation. Pretty sad though.

Anonymous

Anonymous Sun 14 May 2023 19:20:19 No.7375413 Report

Quoted By:

>>7375404
>voldy guide still says to use AOM and similar even though so many are better
that wasn't my point--i wasn't dissing AOM, in fact it does amazing at its job. the better models you speak of are even more overtuned and less generalizable than AOM. all i'm saying is AOM is specifically tuned to make high quality images of girl doing x easier at the cost of not being a more general model like NAI or especially like SD. it's why you don't need a retarded schizo prompt to get a decent image out of it like the latter models i listed.

Anonymous

Anonymous Sun 14 May 2023 19:23:19 No.7375415 Report

Quoted By: >>7375418 >>7375426 >>7375430

>>7375395
Partly this, but I still think that avatarfags and comfyshill being extremely obnoxious completely shunned all non-garbage posters

Anonymous

Anonymous Sun 14 May 2023 19:24:10 No.7375416 Report

Quoted By:

>>7375151
el basado

Anonymous

Anonymous Sun 14 May 2023 19:25:37 No.7375418 Report

Quoted By: >>7375428 >>7375449

>>7375415
i mean yeah that too, but that's just a symptom of newfags joining the general. avatarfags weren't really infesting /sdg/ 6-7 months ago. this place is slowly undergoing the same shit anyways

Anonymous

View Same Google ImgOps iqdb SauceNAO Kotopussy.jpg, 165KiB, 1024x1536

Anonymous Sun 14 May 2023 19:26:19 No.7375421 Report

Quoted By:

>>7375237
>>7375261
Yep, drawn on. Overlay effects would probably disagree with any style loras so it's gotta be a manual thing for now.

Anonymous

Anonymous Sun 14 May 2023 19:28:00 No.7375425 Report

Quoted By:

Has anyone made a toraishi lora?

Anonymous

Anonymous Sun 14 May 2023 19:30:07 No.7375426 Report

Quoted By:

>>7375415
yeah the sheer volume of people posting nothing but one character/style and treating the general as their personal blog is baffling. I get it if you're testing and want feedback, but most of these people are just chasing clout

Anonymous

Anonymous Sun 14 May 2023 19:30:18 No.7375428 Report

Quoted By: >>7375431

>>7375418
I wish I joined earlier myself. I really got into local SD in late February, right when the things pretty much got settled and started to get a bit stale.
Exploring prompts on NAI still felt interesting but kinda vain with all those easy models laying around.

Anonymous

Anonymous Sun 14 May 2023 19:31:20 No.7375430 Report

Quoted By:

>>7375415
this is 90% of why it went to shit, if this place becomes like sdg I'll give up on it too

Anonymous

Anonymous Sun 14 May 2023 19:32:26 No.7375431 Report

Quoted By:

>>7375428
Real niggas remember wrangling VQGAN CLIP

Anonymous

Anonymous Sun 14 May 2023 19:33:17 No.7375434 Report

Quoted By: >>7375611

>>7375088
>
tight black lingerie is the best for thicc killjoy

Anonymous

Anonymous Sun 14 May 2023 19:34:41 No.7375435 Report

Quoted By: >>7375612

I need to put a baby in Thiccjoy

Anonymous

Anonymous Sun 14 May 2023 19:36:54 No.7375438 Report

Quoted By: >>7375440

Whats the trick to areola slip? Tagging it and areolae never seems to work

Anonymous

Anonymous Sun 14 May 2023 19:37:28 No.7375440 Report

Quoted By:

>>7375438
You need to convince the AI that you really really don't want it so it will randomly do it to spite you.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00900-871154-1girl, hair ornamen (...).png, 700KiB, 640x896

Anonymous Sun 14 May 2023 19:38:26 No.7375442 Report

Quoted By: >>7375454

>got randomly flipped off
:(

Anonymous

Anonymous Sun 14 May 2023 19:43:51 No.7375449 Report

Quoted By: >>7375455 >>7375462

>>7375418
>this place is slowly undergoing the same shit anyways
I don't really see it

Anonymous

View Same Google ImgOps iqdb SauceNAO 00252-2878208964.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 19:44:41 No.7375450 Report

Quoted By: >>7375965

Anonymous

Anonymous Sun 14 May 2023 19:53:19 No.7375454 Report

Quoted By: >>7375457

>>7375442
Yaranaika style lora?

Anonymous

Anonymous Sun 14 May 2023 19:54:21 No.7375455 Report

Quoted By:

>>7375449
cool

Anonymous

View Same Google ImgOps iqdb SauceNAO 00034-558571-1girl, hair ornamen (...).png, 2MiB, 1280x1792

Anonymous Sun 14 May 2023 19:59:16 No.7375457 Report

Quoted By: >>7375462

Crimson Comics style lora. Trained to work at 1.0 with my characters.
https://files.catbox.moe/kbg3j9.safetensors
https://rentry.org/gktwz
Man, this style makes me feel things.
>>7375454
Almost.

Anonymous

Anonymous Sun 14 May 2023 19:59:22 No.7375458 Report

Quoted By:

>>7375050
>>7375085
that's why people don't update the pre-gradio version and previous torch

Anonymous

Anonymous Sun 14 May 2023 20:00:43 No.7375462 Report

Quoted By: >>7375464 >>7375465

>>7375449
really? this >>7375457
guy exists, so i definitely see it

Anonymous

Anonymous Sun 14 May 2023 20:00:46 No.7375463 Report

Quoted By: >>7375467 >>7375474

>>7375395
I recently visited /aids/ for the first time in months to see if there were any textgen news and I couldn't recognize the place. Gone where the people sharing their prompt (more of a story start in text gen), replaced by wojak posting, doomposting, just 2 more weeks waitposting.
Really sad, that place used to be such a good general during summer dragon and even early NAI.

Anonymous

Anonymous Sun 14 May 2023 20:01:58 No.7375464 Report

Quoted By:

>>7375462
Yo, I'm a thread personality now?
Damn it.

Anonymous

Anonymous Sun 14 May 2023 20:02:22 No.7375465 Report

Quoted By: >>7375472

>>7375462
No shit there are recognizable "characters" in a general like this, but at least he's somewhat contributing unlike the retards at sdg who just spam same prompt avatars.

Anonymous

Anonymous Sun 14 May 2023 20:02:58 No.7375467 Report

Quoted By:

>>7375463
That's what two years of getting fucked in the ass with no lube does to a general tbf.

Anonymous

Anonymous Sun 14 May 2023 20:05:24 No.7375472 Report

Quoted By: >>7375483

>>7375465
meh, he shits up the thread and spams it everytime he gets angry so i think it'd be better off if he fucked off, but i don't give a shit at this point anyways. nothing actually useful has been posted here in months lol

Anonymous

Anonymous Sun 14 May 2023 20:06:43 No.7375474 Report

Quoted By:

>>7375463
It's still way more healthy and active that is has been in years, no doubt because of normalfag influx after textgen became big again.

Anonymous

Anonymous Sun 14 May 2023 20:12:37 No.7375483 Report

Quoted By:

>>7375472
I don't think we'll see any significant advancements at this point unless there's another model leak.
You can do plenty of shit with what we have today (loras/finetunes/extensions/inpaint etc) if you have a good enough GPU and enough persistence, and we can't really do much with the base model being kinda crap

Anonymous

Anonymous Sun 14 May 2023 20:12:51 No.7375484 Report

Quoted By: >>7375557

>>7375295
Kicking her belly until she miscarriages... kino.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00108-1946687544.png, 1MiB, 960x1344

Anonymous Sun 14 May 2023 20:13:29 No.7375485 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 20:16:46 No.7375487 Report

Quoted By: >>7375489 >>7375493

>>7375395
>the need for experimenting with prompting was never there
been here before textural inversion was ever a thing, im glad that you don't need to go full schizo mode in order to make something bearable out of fucking WD and i don't want newfags to go through that shit aswel. remember a proper tag for "see-through clothes" from SD? me neither, and thats for good

>why would you bother with style prompts when you can just slap on a few loras and use AOM
still you can be doing stuff with (flat colors), (sketch), (traditional media) and many more, and you're saying like combining loras together to make a middle of them is something bad? same experiments but instead of shitty ass reagents you have hundred times better ones

Anonymous

Anonymous Sun 14 May 2023 20:19:17 No.7375489 Report

Quoted By:

>>7375487
>you're saying like combining loras together to make a middle of them is something bad?
where did i say anyone of this was bad? i was explaining why creative ideas and technical prompts aren't a thing anymore for the most part

Anonymous

Anonymous Sun 14 May 2023 20:22:19 No.7375493 Report

Quoted By: >>7375495

>>7375487
Why the fuck did you take that as a personal attack and went full retard? Nowhere that anon said it was a bad thing.

Anonymous

Anonymous Sun 14 May 2023 20:24:02 No.7375495 Report

Quoted By: >>7375517

>>7375493
>personal attack
self projecting much

Anonymous

Anonymous Sun 14 May 2023 20:28:36 No.7375505 Report

Quoted By:

Gonna rebake aom2strawberry with the different AOMs just to see what happens.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00011-438778590.0.png, 852KiB, 1024x1024

Anonymous Sun 14 May 2023 20:36:52 No.7375517 Report

Quoted By:

>>7375495
>self projecting much
>you keep using that word

Anonymous

View Same Google ImgOps iqdb SauceNAO 00319-520372276.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 20:43:11 No.7375523 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 01668-3054820133.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 20:46:27 No.7375526 Report

Quoted By: >>7375528 >>7375529 >>7375798

Anonymous

View Same Google ImgOps iqdb SauceNAO 00324-3694818469.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 20:47:18 No.7375527 Report

Quoted By: >>7375965

Anonymous

Anonymous Sun 14 May 2023 20:47:22 No.7375528 Report

Quoted By:

>>7375526
Nice

Anonymous

View Same Google ImgOps iqdb SauceNAO 01699-4088438267.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 20:47:32 No.7375529 Report

Quoted By:

>>7375526

Anonymous

Anonymous Sun 14 May 2023 20:47:50 No.7375530 Report

Quoted By: >>7375539

>>7373269
Hey guys. So I don't know what the fuck happened but ever since today, it's been getting stuck at 50 fucking percent and just stays there and hangs. No matter what fucking config I set it to, it stops at 50% and the ETA just keeps increasing. The piece of shit even showed it go up to 100% and then went back down to 48% just now and has been sticking there. Does anyone know whats going on?

Anonymous

View Same Google ImgOps iqdb SauceNAO 01716-1245173150.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 20:49:29 No.7375532 Report

Quoted By: >>7375535

Anonymous

View Same Google ImgOps iqdb SauceNAO 01739-502301774.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 20:50:49 No.7375535 Report

Quoted By: >>7375537

>>7375532

Anonymous

View Same Google ImgOps iqdb SauceNAO 01745-2142600682.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 20:51:41 No.7375537 Report

Quoted By:

>>7375535

Anonymous

Anonymous Sun 14 May 2023 20:52:03 No.7375539 Report

Quoted By: >>7375541 >>7375549

>>7375530
Just to give more information: I've tried running it without a vae, upscaler. Hi-res fix is enabled, I havent tried to run it without it yet. I just restarted my computer so I'll see if that fixes it.

Anonymous

Anonymous Sun 14 May 2023 20:52:56 No.7375541 Report

Quoted By: >>7375551

>>7375539
>Hi-res fix is enabled
it's definitely somehow related to this if it's stopping at 50%

Anonymous

Anonymous Sun 14 May 2023 20:53:51 No.7375544 Report

Quoted By:

I sure fucking LOVE when its unclear whether the cloth is a jacket, coat, sweater or a fucking shirt. jesus christ at least im almost done tagging

Anonymous

Anonymous Sun 14 May 2023 20:54:01 No.7375545 Report

Quoted By: >>7375546

>>7375104
Doesn't medvram mess with the output quality somehow?

Anonymous

Anonymous Sun 14 May 2023 20:55:01 No.7375546 Report

Quoted By:

>>7375545
Wrong reply >>7375105

Anonymous

View Same Google ImgOps iqdb SauceNAO 00027-3654938155.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 20:56:09 No.7375549 Report

Quoted By: >>7375551 >>7375555

>>7375539
Wait a minute, are you saying you're using hi-res fix without an upscaler? Does that even work? Did it work for you before?

Anonymous

Anonymous Sun 14 May 2023 20:57:15 No.7375551 Report

Quoted By: >>7375572

>>7375541
>>7375549
False alarm. It seems to be working again after a reboot.

Anonymous

Anonymous Sun 14 May 2023 21:00:11 No.7375555 Report

Quoted By: >>7375572

>>7375549
>Does that even work?
NTa, but actually better than with upscaler, lol.

Anonymous

Anonymous Sun 14 May 2023 21:01:37 No.7375557 Report

Quoted By:

>>7375484
based pregnant woman kicker

Anonymous

Anonymous Sun 14 May 2023 21:11:01 No.7375572 Report

Quoted By: >>7375577

>>7375551
i can't believe no one has told you this but the reason why it was hanging up was some other process was using vram that closed when you restarted your pc. check for that first next time
>>7375555
honestly just stop, that's not even upscaling

Anonymous

Anonymous Sun 14 May 2023 21:12:09 No.7375573 Report

Quoted By: >>7375577

I wanna make a lora with crappy jpgs, should I upscale/noise reduce them or just use them as is?

Anonymous

Anonymous Sun 14 May 2023 21:15:32 No.7375577 Report

Quoted By: >>7375582

>>7375572
>honestly just stop, that's not even upscaling
I could bet you you wouldn't see the difference, but that's too much effort when you're gonna ignore it and call me a faggot anyways.
>>7375573
I upscale.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00001-461091581.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 21:17:33 No.7375580 Report

Quoted By: >>7375591

it seems that "nsfw", "masterpiece" and "best quality" were fucking up my gens alot

Anonymous

View Same Google ImgOps iqdb SauceNAO 00025-4105654901.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 21:20:26 No.7375582 Report

Quoted By: >>7375583 >>7375584

https://mega.nz/folder/ctl1FYoK#BYlpywutnH4psbEcxsbLLg
added enpe
>>7375577
>we now have people trying to insist there is no noticeable difference between ai-assisted upscaling and traditional spatial upscaling
lol /hdg/ never change

Anonymous

View Same Google ImgOps iqdb SauceNAO 00080-4003475690.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 21:21:27 No.7375583 Report

Quoted By:

>>7375582

Anonymous

Anonymous Sun 14 May 2023 21:21:28 No.7375584 Report

Quoted By: >>7375597

>>7375582
>lol /hdg/ never change
I actually retract my statement because I never realized highres fix doesn't work on the same principle as img2img upscaling.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00026-2547202802.png, 848KiB, 1024x1536

Anonymous Sun 14 May 2023 21:23:45 No.7375586 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 21:26:05 No.7375591 Report

Quoted By:

>>7375580
nsfw is somewhat useful but yeah drop best quality and masterpiece

Anonymous

Anonymous Sun 14 May 2023 21:30:43 No.7375597 Report

Quoted By:

>>7375584
(What I mean is that highresfix with no upscaler will just be blurry, while img2img with no upscaler gives you the same quality as highres with upscaler, all while taking the same VRAM.) Weird.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00059-2802436990.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 21:32:04 No.7375598 Report

Quoted By: >>7375620 >>7375627 >>7375664 >>7375678

post kimonos

Anonymous

View Same Google ImgOps iqdb SauceNAO tmpp6n9pc1f.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 21:34:18 No.7375603 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 668fb158a4d3200e3df9aaa23cd5f65f.jpg, 48KiB, 500x340

Anonymous Sun 14 May 2023 21:36:15 No.7375607 Report

Quoted By: >>7375618 >>7375620

When lora training a character how do you organize the training data for different outfits?

would it be seperating each individual outfit and having on dedicated to to character or everythig in one folder.

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765662-2716041678-kill (...).png, 2MiB, 1600x2000

Anonymous Sun 14 May 2023 21:38:19 No.7375611 Report

Quoted By:

>>7375434
thanks. pattern type lingerine are hard ones since consistency of same pattern wont be same when upscaling

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765668-3611219051-nsfw (...).png, 2MiB, 1600x1800

Anonymous Sun 14 May 2023 21:39:07 No.7375612 Report

Quoted By:

>>7375435
pregnancy is not kinda my thing but hope you like creampie instead

Anonymous

Anonymous Sun 14 May 2023 21:40:38 No.7375618 Report

Quoted By: >>7375632

>>7375607
just put them in separate folders, the tagging is whats really important here so you could just put them all in one folder and itd work fine but having multiple folders makes organization much easier and it allows you to adjust repeats for outfits that are less represented in your dataset

Anonymous

View Same Google ImgOps iqdb SauceNAO 22003-2704538299-1girl, solo, re (...).png, 925KiB, 768x1152

Anonymous Sun 14 May 2023 21:41:43 No.7375620 Report

Quoted By: >>7375632

>>7375598
>

>>7375607
Personally I don't bother with separating them although to be fair I don't really deal with characters who tend to have multiple or varied outfits in their fanart in the first place
Technically speaking separating the images into subfolders shouldn't be necessary as long as the various outfits and their components are tagged appropriately. The main purpose of subfolders is in case you want certain images to have more repeats compared to the rest of the set

Anonymous

View Same Google ImgOps iqdb SauceNAO 00332-2578035218.png, 2MiB, 1280x1536

Anonymous Sun 14 May 2023 21:42:56 No.7375623 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00053-2737296622-1girl, hair orn (...).png, 2MiB, 1280x1792

Anonymous Sun 14 May 2023 21:45:56 No.7375627 Report

Quoted By:

>>7375598
Ok.

Anonymous

Anonymous Sun 14 May 2023 21:49:16 No.7375632 Report

Quoted By: >>7375635 >>7375670

>>7375618
If i were to go with separating each of the outfit, i will just tag them approprately then adjust the repeats so it appears thats if i understand.

Is there a way to use an activation word for that particular outfit say if two outfits have similar dress such as a t shirt and pants so doesn't confuse the two in tagging.

>>7375620
I see it doesnt make quite a difference whether the outfits are seperated or not so long as the tagging is correct.

Is it possible to see a dataset example and how to go about the different tags that activate those particular outfits.

Anonymous

Anonymous Sun 14 May 2023 21:51:08 No.7375635 Report

Quoted By:

>>7375632
You can view the metadata of loras (provided they aren't ancient) in the webui to see their tags

Anonymous

View Same Google ImgOps iqdb SauceNAO file.png, 1MiB, 1024x1024

Anonymous Sun 14 May 2023 21:51:30 No.7375637 Report

Quoted By: >>7375646 >>7376018

This panty reveal

Anonymous

View Same Google ImgOps iqdb SauceNAO AlearWish.png, 1MiB, 856x1192

Anonymous Sun 14 May 2023 21:52:30 No.7375641 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO Alear'sBackstory.png, 2MiB, 1200x1800

Anonymous Sun 14 May 2023 21:53:03 No.7375643 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 21:57:04 No.7375646 Report

Quoted By:

>>7375637
sasuga Hinata chan

Anonymous

Anonymous Sun 14 May 2023 22:00:05 No.7375649 Report

Quoted By: >>7375662

i have a RTX 3080Ti (12 GB VRAM), and i keep getting out of memory errors when i try to generate anything above 700x700 with an upscale of 1.2/3, is there anything i can do about it?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00071-1448014447-1girl, hair orn (...).png, 2MiB, 1792x1280

Anonymous Sun 14 May 2023 22:04:53 No.7375656 Report

Quoted By:

One day, landscapebros.

Anonymous

Anonymous Sun 14 May 2023 22:09:20 No.7375662 Report

Quoted By: >>7375734

>>7375649
--half mem or something similar added to your command_args

Anonymous

View Same Google ImgOps iqdb SauceNAO 9996490989765670-3017006550-kill (...).png, 3MiB, 1600x2400

Anonymous Sun 14 May 2023 22:11:46 No.7375664 Report

Quoted By:

>>7375598
quick one before going to sleep

Anonymous

View Same Google ImgOps iqdb SauceNAO 00003-2519438140.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 22:13:20 No.7375669 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 22:13:41 No.7375670 Report

Quoted By:

>>7375632
>Is there a way to use an activation word for that particular outfit say if two outfits have similar dress such as a t shirt and pants so doesn't confuse the two in tagging.
you can use an unique keyword to distinguish the 2 and just add them to all the appropriate images, since you're training the text encoder as well during LoRA training something like that will work fine. It will eventually pick up on the fact that a specific token constantly shows up each time on a specific outfit and start associating them

Anonymous

View Same Google ImgOps iqdb SauceNAO 00011-1758193733.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 22:30:11 No.7375678 Report

Quoted By:

>>7375598
If we're going to post kimonos on /h/, we might as well make them /h/-relevant.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00007-1711569305.png, 3MiB, 1280x1920

Anonymous Sun 14 May 2023 22:30:33 No.7375679 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 22:33:01 No.7375682 Report

Quoted By: >>7375685 >>7375696

Are reference sheets good for loras? Assuming it's tagged correctly. Or is it better to crop each part into a separate image?

Anonymous

Anonymous Sun 14 May 2023 22:33:49 No.7375685 Report

Quoted By:

>>7375682
Should be fine

Anonymous

Anonymous Sun 14 May 2023 22:38:49 No.7375692 Report

Quoted By: >>7375706

Still on a9fed, do I even bother pulling?

Anonymous

View Same Google ImgOps iqdb SauceNAO 01622-2769430205.png, 1MiB, 1024x1536

Anonymous Sun 14 May 2023 22:39:58 No.7375696 Report

Quoted By: >>7375704

>>7375682
If you don't crop each part into a separate image, make sure you include the "multiple views" tag.

Anonymous

Anonymous Sun 14 May 2023 22:43:27 No.7375704 Report

Quoted By: >>7375719 >>7375722

>>7375696
Should I just use the distinct views of the whole body or do random little details of the outfit/etc help/hurt it?

Anonymous

Anonymous Sun 14 May 2023 22:44:30 No.7375706 Report

Quoted By:

>>7375692
Fuck no.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00015-1776457389.png, 3MiB, 1280x1920

Anonymous Sun 14 May 2023 22:45:23 No.7375709 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00012-1776457386.png, 3MiB, 1280x1920

Anonymous Sun 14 May 2023 22:47:39 No.7375711 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00233-771677354-nsfw, scan, (ske (...).png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 22:51:19 No.7375718 Report

Quoted By: >>7375806

Anonymous

View Same Google ImgOps iqdb SauceNAO 00032-2785892688.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 22:52:47 No.7375719 Report

Quoted By: >>7375722

Weeb elves are best elves.

>>7375704
I don't know, the effects could be completely random. Tag everything that appears in each view, and maybe the trainer won't get confused if you use both "full body" and "close-up" in the same image.

Anonymous

Anonymous Sun 14 May 2023 22:54:19 No.7375722 Report

Quoted By:

>>7375704
>>7375719
just tag it as multiple views and reference sheet then be done with it. you're asking for problems tagging it stuff like full body and close-up

Anonymous

View Same Google ImgOps iqdb SauceNAO 34234.jpg, 37KiB, 555x124

Anonymous Sun 14 May 2023 22:54:32 No.7375723 Report

Quoted By: >>7375725 >>7375753 >>7375755 >>7375756

>got top 50 image on pixiv
What do I win?

Anonymous

Anonymous Sun 14 May 2023 22:55:24 No.7375725 Report

Quoted By: >>7375728 >>7375729 >>7375744

>>7375723
ding dong bannu

Anonymous

Anonymous Sun 14 May 2023 22:56:17 No.7375728 Report

Quoted By:

>>7375725
Did they extend that ban to the whole site or is it still just pixiv? It's fun watching number go up but I mostly just gen one artist.

Anonymous

Anonymous Sun 14 May 2023 22:56:55 No.7375729 Report

Quoted By: >>7375756

>>7375725
I've already been warned like 5 times for not enough censorship. I don't know what it takes to get a ban.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_5syp86.png, 1MiB, 832x1184

Anonymous Sun 14 May 2023 22:57:56 No.7375733 Report

Quoted By: >>7375741 >>7375812

Kimono time?

Anonymous

Anonymous Sun 14 May 2023 22:58:04 No.7375734 Report

Quoted By:

>>7375662
--medvram ?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00246-1562920935-nsfw, scan, (sk (...).png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 23:02:55 No.7375741 Report

Quoted By: >>7375748 >>7375812

>>7375733
Yes but also undressing.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00011-827812807.jpg, 324KiB, 2304x2304

Anonymous Sun 14 May 2023 23:03:32 No.7375743 Report

Quoted By:

Anonymous

Anonymous Sun 14 May 2023 23:03:44 No.7375744 Report

Quoted By: >>7375756

>>7375725
I don't use fanbox or requests. Haven't made a cent off my stuff. Just do it for fun.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_kw3o2q.png, 1MiB, 832x1184

Anonymous Sun 14 May 2023 23:05:28 No.7375748 Report

Quoted By: >>7375771

>>7375741
Shes good at that part

Anonymous

Anonymous Sun 14 May 2023 23:07:08 No.7375753 Report

Quoted By:

>>7375723
you got enough internet attention
leave us alone

Anonymous

View Same Google ImgOps iqdb SauceNAO 00879-161108346.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 23:08:33 No.7375755 Report

Quoted By: >>7375767

>>7375723
I can't tell from looking at the ranking chart which one was yours. Did you tag it as AI?

Anonymous

Anonymous Sun 14 May 2023 23:08:38 No.7375756 Report

Quoted By:

>>7375723
>>7375729
>>7375744
unironically: no one asked

Anonymous

Anonymous Sun 14 May 2023 23:08:55 No.7375758 Report

Quoted By:

>Characters, styles, concepts
Objects use the same training rules as characters, right?

Anonymous

Anonymous Sun 14 May 2023 23:12:56 No.7375763 Report

Quoted By: >>7375766 >>7375796

What model should I try instead of the NAI one?

Anonymous

Anonymous Sun 14 May 2023 23:14:12 No.7375766 Report

Quoted By:

>>7375763
wd1.3

Anonymous

Anonymous Sun 14 May 2023 23:14:42 No.7375767 Report

Quoted By: >>7375794

>>7375755
>Did you tag it as AI?
Yes but I won't post it here because I don't want to shill.
also
>check which image is the top ai image for the day
>literal nai generic image
>check tags
>fotm chinese gacha game but the characters in the image look nothing like those characters other than hair color
I will never understand.

Anonymous

Anonymous Sun 14 May 2023 23:16:07 No.7375769 Report

Quoted By: >>7375773

>Want a gen of a girl sitting on a throne
>It's actually kind of a struggle
I swear it wasn't like this back in NAI

Anonymous

View Same Google ImgOps iqdb SauceNAO 00339-953923420.png, 2MiB, 1536x1280

Anonymous Sun 14 May 2023 23:17:23 No.7375770 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 00287-3490585317-nsfw, scan, (sk (...).png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 23:17:49 No.7375771 Report

Quoted By: >>7375803

>>7375748
Very useful skill.

Anonymous

Anonymous Sun 14 May 2023 23:18:11 No.7375773 Report

Quoted By: >>7375780

>>7375769
This is what happens when you use models that output naked ladies when prompting for fettucini.

Anonymous

Anonymous Sun 14 May 2023 23:22:25 No.7375780 Report

Quoted By: >>7375792

>>7375773
I'm gonna prompt my dick in your ass.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00344-4212595060.png, 2MiB, 1536x1280

Anonymous Sun 14 May 2023 23:25:41 No.7375786 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO 1681270683223455.png, 2MiB, 896x1280

Anonymous Sun 14 May 2023 23:27:18 No.7375788 Report

Quoted By: >>7375801 >>7375837

Added fei_(maidoll) lora to the rentry.

https://rentry.org/4ewit

Anonymous

Anonymous Sun 14 May 2023 23:29:47 No.7375792 Report

Quoted By:

>>7375780
Take me out to dinner, first. Vegetarian.

Anonymous

View Same Google ImgOps iqdb SauceNAO gagging is prohibited.jpg, 66KiB, 512x768

Anonymous Sun 14 May 2023 23:30:36 No.7375794 Report

Quoted By: >>7375824

>>7375767
yeah the most activity appears to come from a combination of existing followers to drive views/likes/bookmarks and trending tags.
Tons of starrail images in top 50. If you want to game the system you could probably just gen images using loras that follow trending tags.
that's pretty boring but if you're taking money/requests you'd be foolish not to market yourself in this way.
I throw my stuff on pixiv because it's the least bullshit way to put content online. I tried various boorus including aibooru but my images kept getting unlisted due to a lack of approval and I just didn't see the point in putting in all that effort to tag and source images when they could end up being filtered by a select group simply not liking them enough.
with pixiv you can upload pretty much anything you want so long as you follow their censorship laws, and even then a little blur or mosaic is a lot less effort.

Anonymous

Anonymous Sun 14 May 2023 23:31:57 No.7375796 Report

Quoted By:

>>7375763
aom2 for separated sfw/nsfw, based65 for all in one.

Anonymous

Anonymous Sun 14 May 2023 23:32:05 No.7375798 Report

Quoted By: >>7375832

>>7375526
Which lora did you use for the pose?

Anonymous

Anonymous Sun 14 May 2023 23:34:24 No.7375801 Report

Quoted By:

>>7375788
based

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_zohf90.png, 1MiB, 832x1184

Anonymous Sun 14 May 2023 23:35:49 No.7375803 Report

Quoted By:

>>7375771
Indeed

Anonymous

View Same Google ImgOps iqdb SauceNAO 00138-1101365369.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 23:36:17 No.7375804 Report

Quoted By: >>7375825

Anonymous

Anonymous Sun 14 May 2023 23:36:48 No.7375806 Report

Quoted By:

>>7375718
Catbox?

Anonymous

View Same Google ImgOps iqdb SauceNAO 22068-3692401128-1girl, (red kim (...).png, 1012KiB, 768x1152

Anonymous Sun 14 May 2023 23:39:32 No.7375812 Report

Quoted By:

>>7375733
>>7375741

Anonymous

Anonymous Sun 14 May 2023 23:43:27 No.7375824 Report

Quoted By:

>>7375794
catbox?

Anonymous

Anonymous Sun 14 May 2023 23:43:30 No.7375825 Report

Quoted By: >>7375830

>>7375804
what art style is this

Anonymous

View Same Google ImgOps iqdb SauceNAO 00151-3903905674.png, 2MiB, 1024x1536

Anonymous Sun 14 May 2023 23:46:40 No.7375830 Report

Quoted By:

>>7375825
It's the style LoRA. It got posted ITT earlier. CTRL + F "massakasama"

Anonymous

View Same Google ImgOps iqdb SauceNAO 01697-217331728.png, 2MiB, 1536x1024

Anonymous Sun 14 May 2023 23:47:03 No.7375832 Report

Quoted By: >>7375857

>>7375798
The corneo-cowgirl embedding from Civitai, but I think that has more of an effect on the dick than the girl's pose.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_bdc0pc.png, 1MiB, 832x1184

Anonymous Sun 14 May 2023 23:51:10 No.7375837 Report

Quoted By: >>7375849

>>7375788
Neat

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_pvqgdy.png, 1MiB, 832x1184

Anonymous Mon 15 May 2023 00:01:09 No.7375849 Report

Quoted By:

>>7375837

Anonymous

View Same Google ImgOps iqdb SauceNAO 1668896861432326.png, 1MiB, 1280x896

Anonymous Mon 15 May 2023 00:08:39 No.7375851 Report

Quoted By:

Anonymous

Anonymous Mon 15 May 2023 00:11:28 No.7375855 Report

Quoted By: >>7375867 >>7375888

>>CUDA out of memory. Tried to allocate 640.00 MiB (GPU 0; 8.00 GiB total capacity; 5.69 GiB already allocated; 0 bytes free; 7.25 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
what the fug

Anonymous

Anonymous Mon 15 May 2023 00:14:51 No.7375857 Report

Quoted By:

>>7375832
Neat, thanks for the info

Anonymous

View Same Google ImgOps iqdb SauceNAO 00160-2195473148.png, 3MiB, 1792x1080

Anonymous Mon 15 May 2023 00:17:47 No.7375860 Report

Quoted By: >>7375862 >>7375885 >>7375903 >>7376026

>https://huggingface.co/waifu-diffusion/wd-1-5-beta3

Anonymous

Anonymous Mon 15 May 2023 00:21:20 No.7375862 Report

Quoted By: >>7376032

>>7375860
>The WD 1.5 Base model is only intended for training use. For generation, it is recomended to create your own finetunes and loras on top of WD 1.5 Base or use one of the aesthetic models. More information and sample generations for the aesthetic models are in the release notes

Anonymous

Anonymous Mon 15 May 2023 00:25:51 No.7375867 Report

Quoted By: >>7375878 >>7375884

>>7375855
add this line to your webui-user
>set PYTORCH_CUDA_ALLOC_CONF=garbage_collection_threshold:0.9,max_split_size_mb:512
also if you haven't done that already, add these to commandline args
>--medvram --xformers --always-batch-cond-uncond

Anonymous

View Same Google ImgOps iqdb SauceNAO candy.png, 146KiB, 968x968

Anonymous Mon 15 May 2023 00:30:22 No.7375875 Report

Quoted By: >>7375890

Anonymous

Anonymous Mon 15 May 2023 00:30:56 No.7375878 Report

Quoted By:

>>7375867
i'll do that, thanks.

Anonymous

Anonymous Mon 15 May 2023 00:35:56 No.7375884 Report

Quoted By: >>7375888

>>7375867
What's that first one? Haven't ever seen it before.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00056-457697208-tatsumaki, curly (...).png, 2MiB, 1280x1792

Anonymous Mon 15 May 2023 00:36:26 No.7375885 Report

Quoted By:

>>7375860
bruh why does it still look shit after another 2(?) months of training

Anonymous

Anonymous Mon 15 May 2023 00:38:20 No.7375888 Report

Quoted By:

>>7375884
literally in the error message of >>7375855
makes it gc more often and allocate smaller portions of vram at a time

Anonymous

Anonymous Mon 15 May 2023 00:40:00 No.7375889 Report

Quoted By: >>7375894 >>7375905 >>7375906

>recall anon saying furry models are lightyears ahead of our weebshit
>try some out, out of curiosity
>its NAI tier

what the fuck

Anonymous

Anonymous Mon 15 May 2023 00:41:36 No.7375890 Report

Quoted By: >>7375931

>>7375875
based. catbox?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00002-1457045727.png, 2MiB, 1440x1440

Anonymous Mon 15 May 2023 00:42:35 No.7375892 Report

Quoted By:

Anonymous

Anonymous Mon 15 May 2023 00:43:40 No.7375894 Report

Quoted By: >>7375899 >>7375928

>>7375889
not the ones on civitai

Anonymous

Anonymous Mon 15 May 2023 00:45:05 No.7375899 Report

Quoted By:

>>7375894
kek

Anonymous

View Same Google ImgOps iqdb SauceNAO image.png, 28KiB, 939x190

Anonymous Mon 15 May 2023 00:47:52 No.7375903 Report

Quoted By: >>7375999

>>7375860
>training for months
>"oh shit we weren't training the text encoder"
based retards

Anonymous

View Same Google ImgOps iqdb SauceNAO 01641-248145212.png, 2MiB, 1536x1024

Anonymous Mon 15 May 2023 00:49:34 No.7375905 Report

Quoted By: >>7375913 >>7375928 >>7375971

>>7375889
The furry models that /trash/ prefers to use include all artist tags from the e621 dataset, so you're supposed to add the name of one or more furry artists to the prompt for best results.

Anonymous

Anonymous Mon 15 May 2023 00:50:09 No.7375906 Report

Quoted By:

>>7375889
Same experience.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_7mhugj.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 00:50:59 No.7375909 Report

Quoted By: >>7375925

Anonymous

Anonymous Mon 15 May 2023 00:57:24 No.7375913 Report

Quoted By:

>>7375905
I forget which they were, but one of their hacks is to put a couple of the most godawful artists on there into their negatives

Anonymous

Anonymous Mon 15 May 2023 00:59:44 No.7375921 Report

Quoted By: >>7375922 >>7375926

wtf I spent all that time cropping and resizing images but you don't have to anymore? does resizing help or hurt?

Anonymous

Anonymous Mon 15 May 2023 01:01:23 No.7375922 Report

Quoted By:

>>7375921
It does virtually nothing
You'll get a more noticeable effect from fucking up a tag than cropping
No you don't need to crop for buckets either, it'll do it all automatically, no matter the sizes

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_f07wi7.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 01:02:23 No.7375925 Report

Quoted By:

>>7375909

Anonymous

Anonymous Mon 15 May 2023 01:02:47 No.7375926 Report

Quoted By:

>>7375921
Using latent upscale on really tiny images for added detail might help a little but otherwise cropping and resizing is either automated or obsolete now

Anonymous

Anonymous Mon 15 May 2023 01:04:55 No.7375928 Report

Quoted By: >>7375948 >>7375971

>>7375905
Tried that with yiffymix off of civshit >>7375894

Still looks god awful, no clue what these furries are on about

Anonymous

View Same Google ImgOps iqdb SauceNAO spread.png, 74KiB, 242x242

Anonymous Mon 15 May 2023 01:08:13 No.7375931 Report

Quoted By:

>>7375890
https://files.catbox.moe/efep42.txt
I recommend blackboard lora

Anonymous

View Same Google ImgOps iqdb SauceNAO 00020-1270712887.png, 3MiB, 1440x1440

Anonymous Mon 15 May 2023 01:13:21 No.7375936 Report

Quoted By: >>7376027

Anonymous

View Same Google ImgOps iqdb SauceNAO lora.jpg, 49KiB, 959x387

Anonymous Mon 15 May 2023 01:22:52 No.7375944 Report

Quoted By: >>7375949 >>7375974

alright I'm ready to fuck up in the kitchen but my oven won't turn on what do I do

Anonymous

View Same Google ImgOps iqdb SauceNAO 01868-4215669284.png, 2MiB, 1536x1024

Anonymous Mon 15 May 2023 01:24:31 No.7375948 Report

Quoted By: >>7375961 >>7375971

>>7375928
Fluffyrock is the real leading-edge furry model. New epochs are trained about once a week:
https://huggingface.co/lodestones/furryrock-model-safetensors/tree/main/fluffyrock-1088-megares

Anonymous

Anonymous Mon 15 May 2023 01:25:27 No.7375949 Report

Quoted By:

>>7375944
It's too dark. The model can't see.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00361-2169257846.png, 2MiB, 1536x1280

Anonymous Mon 15 May 2023 01:30:30 No.7375951 Report

Quoted By:

Anonymous

Anonymous Mon 15 May 2023 01:31:07 No.7375952 Report

Quoted By:

When i use controlnet to replicate the pose portrayed in a picture (but with a different character), do i need to add the specifics of the pose in the prompt? Or are the character's tags enough? I am using softedge HED

Anonymous

Anonymous Mon 15 May 2023 01:43:38 No.7375961 Report

Quoted By: >>7375963

>>7375948
Tried this out, better than the civit crap but I still dont get the hype

Anonymous

Anonymous Mon 15 May 2023 01:45:58 No.7375963 Report

Quoted By:

>>7375961
Eh, if all else fails ask /trash/ for help.

Anonymous

Anonymous Mon 15 May 2023 01:47:20 No.7375965 Report

Quoted By:

>>7375450
>>7375527
catbox?

Anonymous

Anonymous Mon 15 May 2023 02:02:47 No.7375971 Report

Quoted By: >>7375972 >>7375975

>>7375948
>>7375928
>>7375905
could those models do this?
https://exhentai.org/s/16eb382162/993792-18

no really a furry or zoophile,
its more of a masochistic humiliation thing for me.

Anonymous

Anonymous Mon 15 May 2023 02:03:22 No.7375972 Report

Quoted By:

>>7375971
ask trash bro why are you here

Anonymous

View Same Google ImgOps iqdb SauceNAO lora2.jpg, 40KiB, 857x275

Anonymous Mon 15 May 2023 02:05:03 No.7375974 Report

Quoted By:

>>7375944
made some progress(?) but stumped again, what do

Anonymous

Anonymous Mon 15 May 2023 02:05:29 No.7375975 Report

Quoted By: >>7376170

>>7375971
speaking of, i don't see many footfags here,
is it cause they are not here, or cause models just not good at feet?

Anonymous

Anonymous Mon 15 May 2023 02:10:57 No.7375981 Report

Quoted By: >>7375984

i have a dataset comprised of around 50% game screenshots and around 50% fan art. all of the screenshots were captioned as 3d and all of the artwork was captioned as 2d.

i can train the LORA past the point where it overfits on the screenshots and then just use something like (2d:1.3) in the prompt and negate most of the style without losing much of the accuracy, or throw 3d in the neg to completely negate the style, albeit at reduced accuracy

picrel, left to right is 3d positive, 2d positive, 2d positive and 3d negative, and neither with the LORA deactivated. the positive 2d one retains the original style the most i think.

i know that people itt have been wanting to train on VN sprites before, something similar could probably be applied there

Anonymous

View Same Google ImgOps iqdb SauceNAO asdfdfxc.jpg, 1MiB, 4096x1536

Anonymous Mon 15 May 2023 02:12:07 No.7375984 Report

Quoted By:

>>7375981
idk how i manage to do this every time

Anonymous

Anonymous Mon 15 May 2023 02:14:16 No.7375988 Report

Quoted By: >>7375996 >>7376013

I have seen a few Japanese that beg for Lora to make more art in there style of his OC. I have also ran into one that rallied his community to cancel a guy that made a Lora of him, How do you guys feel about artist Lora to me as long as it's being used for fun not profit I'm out with my art being used.

Anonymous

Anonymous Mon 15 May 2023 02:18:49 No.7375996 Report

Quoted By: >>7376033

>>7375988
i dont understand anything you are saying.

Anonymous

Anonymous Mon 15 May 2023 02:21:45 No.7375999 Report

Quoted By:

>>7375903
jesus christ, lmao

Anonymous

Anonymous Mon 15 May 2023 02:36:46 No.7376006 Report

Quoted By:

has there been any new VAEs trained??

Anonymous

View Same Google ImgOps iqdb SauceNAO 00807-3474249713.png, 2MiB, 1120x1568

Anonymous Mon 15 May 2023 02:42:41 No.7376013 Report

Quoted By:

>>7375988
I don't really see an issue with artist loras as long as someone isn't charging people money for the lora or trying to sell the generated pictures.
It's something that artists are going to have to adapt to because the simple fact is that if someone can collect 30-50 images of their work, then it's possible to train a lora off of it.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00397-1534922364.png, 2MiB, 1280x1536

Anonymous Mon 15 May 2023 02:43:10 No.7376014 Report

Quoted By:

Anonymous

Anonymous Mon 15 May 2023 02:55:21 No.7376018 Report

Quoted By:

>>7375637
You got owned.

Anonymous

Anonymous Mon 15 May 2023 03:02:14 No.7376023 Report

Quoted By: >>7376024

you know the reason that the keys of english keyboards are arranged the way they are was that back in the days of type writers they had to slow peoples typing so they wouldn't break them so they arranged the characters in a way that was hard to type at the time.
Im started to think that civitais web devs are trying to do the same thing.

Anonymous

Anonymous Mon 15 May 2023 03:03:17 No.7376024 Report

Quoted By:

>>7376023
how so?

Anonymous

Anonymous Mon 15 May 2023 03:05:04 No.7376026 Report

Quoted By:

>>7375860
well, because they fixed their text encoder training bug it can actually render popular characters like Miku and Reimu a ton better now. unfortunately they should have figured out this shit months ago.

Anonymous

Anonymous Mon 15 May 2023 03:07:44 No.7376027 Report

Quoted By:

>>7375936
uoh

Anonymous

Anonymous Mon 15 May 2023 03:11:23 No.7376032 Report

Quoted By:

>>7375862
>The WD 1.5 Base model is only intended for training use
something like this is kind of what we need

Anonymous

Anonymous Mon 15 May 2023 03:11:39 No.7376033 Report

Quoted By: >>7376034 >>7376039

>>7375996
Sorry I had a stroke. I saw a Japanese artist asking for a Lora of his Original Character to be made by fans and for them to send him AI fan art. I also witness the next day an artist send his fans to cancel someone sharing a Lora trained on his own Original Character. I am an artist but sorta retired so I have no dog in this fight. I was just curious for opinions on these kinds of scenarios?

Anonymous

Anonymous Mon 15 May 2023 03:13:38 No.7376034 Report

Quoted By:

>>7376033
>I saw a Japanese artist asking for a Lora of his Original Character to be made by fans and for them to send him AI fan art. I also witness the next day an artist send his fans to cancel someone sharing a Lora trained on his own Original Character.
I mean they did what he asked, so what was he expecting?

Anonymous

Anonymous Mon 15 May 2023 03:15:41 No.7376035 Report

Quoted By: >>7376047

which blocks do you keep just for character loras?

Anonymous

Anonymous Mon 15 May 2023 03:16:17 No.7376036 Report

Quoted By:

Uhh how do I merge a lora into a model again?

Anonymous

Anonymous Mon 15 May 2023 03:16:39 No.7376038 Report

Quoted By: >>7376045 >>7376070

new/updated sampler might be coming soon https://github.com/crowsonkb/k-diffusion/issues/59#issuecomment-1547039484

Anonymous

Anonymous Mon 15 May 2023 03:17:20 No.7376039 Report

Quoted By: >>7376043

>>7376033
i have trained a lora of an artist whos patreon im subscribed to,
never released it or any pictures from it, since i dont want to do anything that would discourage him from continueing to do what he does .

Anonymous

Anonymous Mon 15 May 2023 03:22:06 No.7376043 Report

Quoted By: >>7376044

>>7376039
That makes sense to me. I have been subbed to a guy on fanbox for 5 years that hasn't consistently uploaded for nearly 2 years. I have been considering making a Lora for him just cause I miss his art.

Anonymous

Anonymous Mon 15 May 2023 03:24:30 No.7376044 Report

Quoted By:

>>7376043
well the one im sub is the only patreon i m in that has consistently output material, and has been doing so for years

Anonymous

Anonymous Mon 15 May 2023 03:25:14 No.7376045 Report

Quoted By:

>>7376038
yeah but how well does it do it for anime?

Anonymous

Anonymous Mon 15 May 2023 03:25:57 No.7376047 Report

Quoted By: >>7376048

>>7376035
I've been doing output only. Seems to have no real net benefit or negative effect on the LORA, but it trains faster and the output is smaller so why not.

Anonymous

Anonymous Mon 15 May 2023 03:26:51 No.7376048 Report

Quoted By: >>7376049

>>7376047
>output
is output "down", or "up", cause thats how im seeing the labels on the easyscripts?

Anonymous

Anonymous Mon 15 May 2023 03:27:40 No.7376049 Report

Quoted By:

>>7376048
i think it calls the out layers up

Anonymous

View Same Google ImgOps iqdb SauceNAO 01914-3820067218.png, 2MiB, 1536x1024

Anonymous Mon 15 May 2023 03:30:27 No.7376053 Report

Quoted By:

These specimens need to be handled very carefully.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_m6b8e3.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 03:39:01 No.7376061 Report

Quoted By: >>7376069

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_48f52h.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 03:48:52 No.7376069 Report

Quoted By:

>>7376061

Anonymous

View Same Google ImgOps iqdb SauceNAO 00439-4276066846.png, 2MiB, 1280x1536

Anonymous Mon 15 May 2023 03:51:24 No.7376070 Report

Quoted By: >>7376072

>>7376038
As DPM++ SDE/DPM++ SDE Karras only user, thanks for this anon, applied the change and it seems to looks better. And it works on anime models.

Anonymous

Anonymous Mon 15 May 2023 03:54:31 No.7376072 Report

Quoted By: >>7376077

>>7376070
When I tried applying the change it just cooked my outputs, even when I then tried enabling the last sigma being discarded. I must've done something wrong. Nice to see it working for you tho

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_ggrm5d.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 03:54:40 No.7376073 Report

Quoted By: >>7376082

How do you feel about giant elf udders?

Anonymous

Anonymous Mon 15 May 2023 03:58:36 No.7376076 Report

Quoted By:

updates webui to install lycoris, now can't seem to launch through launch.py, can only do so through webui.py
problem is launching through launch.py is how i get my commandline args and i cant function without xformers and medvram
any workaround?

Anonymous

View Same Google ImgOps iqdb SauceNAO 00447-3313008574.png, 2MiB, 1280x1536

Anonymous Mon 15 May 2023 03:59:03 No.7376077 Report

Quoted By: >>7376079

>>7376072
How did you apply? Apply this https://gist.github.com/crowsonkb/3ed16fba35c73ece7cf4b9a2095f2b78 into stable-diffusion-webui\repositories\k-diffusion\k_diffusion\sampling.py

In line 542

Anonymous

Anonymous Mon 15 May 2023 04:01:14 No.7376079 Report

Quoted By: >>7376080

>>7376077
Are you even sure you applied it correctly then? That adds the function to the kdiffusion repo but you then need to reference that new function somewhere in the webui's sd_samplers_kdiffusion module.

Anonymous

Anonymous Mon 15 May 2023 04:02:20 No.7376080 Report

Quoted By: >>7376093

>>7376079
Oh wait no I see what you did, you just overwrote the old sampler. I see the problem as to why my impl didn't work now either, because the webui does have a special case for that one.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_75mcag.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:03:24 No.7376082 Report

Quoted By: >>7376087

>>7376073

Anonymous

Anonymous Mon 15 May 2023 04:06:46 No.7376084 Report

Quoted By: >>7376086

god damn controlnet author on a roll with this shit https://github.com/Mikubill/sd-webui-controlnet/pull/1278

Anonymous

Anonymous Mon 15 May 2023 04:07:55 No.7376086 Report

Quoted By: >>7376097 >>7376109

>>7376084
>https://github.com/Mikubill/sd-webui-controlnet/pull/1278
what does this do?

Anonymous

Anonymous Mon 15 May 2023 04:07:58 No.7376087 Report

Quoted By: >>7376088

>>7376082
cumhat

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_hy9rrg.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:08:32 No.7376088 Report

Quoted By: >>7376101

>>7376087
whoops

Anonymous

View Same Google ImgOps iqdb SauceNAO 00021-2236499419.png, 2MiB, 1280x1536

Anonymous Mon 15 May 2023 04:11:05 No.7376093 Report

Quoted By: >>7376104

>>7376080
Yep, just replaced the sampler for now. You can add another one and then on stable-diffusion-webui\modules\sd_samplers_kdiffusion.py, add a sampler redirecting to that model.

samplers_k_diffusion = [
...

Anonymous

View Same Google ImgOps iqdb SauceNAO bf740c10634ffade26456330af1add01 (...).png, 2MiB, 1024x1536

Anonymous Mon 15 May 2023 04:12:49 No.7376097 Report

Quoted By: >>7376109

>>7376086
Dunno exactly yet, I'm sure he'll drop some documentation soon if it works out. Pic unrelated.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_6ohwqv.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:15:23 No.7376101 Report

Quoted By: >>7376112

>>7376088

Anonymous

View Same Google ImgOps iqdb SauceNAO tmpqwmu3bdb.png, 2MiB, 1536x1024

Anonymous Mon 15 May 2023 04:16:56 No.7376104 Report

Quoted By: >>7376110

>>7376093
kazusa enjoyer...

Anonymous

Anonymous Mon 15 May 2023 04:21:15 No.7376109 Report

Quoted By:

>>7376086
>>7376097
Adaptive instance normalization (AdaIN) is another kind of 1-image style transfer. This is the paper on it:
https://arxiv.org/pdf/1703.06868.pdf

Anonymous

View Same Google ImgOps iqdb SauceNAO 00019-2466976318.png, 2MiB, 1280x1536

Anonymous Mon 15 May 2023 04:24:01 No.7376110 Report

Quoted By:

>>7376104
I trained a lot of BA characters, though the halos sometimes are perfect and sometimes not (gacha gen)

I don't inpaint though, but probably doing that I would get perfect halos everytime.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_qb2npk.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:24:42 No.7376112 Report

Quoted By: >>7376113

>>7376101
god damn I love elves bros

Anonymous

View Same Google ImgOps iqdb SauceNAO 01980-1171767886.png, 2MiB, 1536x1024

Anonymous Mon 15 May 2023 04:27:22 No.7376113 Report

Quoted By: >>7376116

>>7376112
I like my elves to be on alien planets.

Anonymous

View Same Google ImgOps iqdb SauceNAO 00388-3527121256 (1).png, 1MiB, 1024x1536

Anonymous Mon 15 May 2023 04:30:47 No.7376114 Report

Quoted By:

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_oujgof.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:32:17 No.7376116 Report

Quoted By: >>7376119

>>7376113
Would you like them in a house?
Would you like them with a mouse?

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_0vkd1z.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:37:02 No.7376119 Report

Quoted By: >>7376129

>>7376116

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_it8xr5.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:48:17 No.7376129 Report

Quoted By: >>7376133

>>7376119
I think making Killjoy an elf tries to turn her beanie into a santa hat

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_lpmtjy.png, 1MiB, 1024x1024

Anonymous Mon 15 May 2023 04:52:57 No.7376133 Report

Quoted By:

>>7376129

Anonymous

Anonymous Mon 15 May 2023 05:12:10 No.7376138 Report

Quoted By:

>>7376137
>>7376137
>>7376137

move it coomers

Anonymous

Anonymous Mon 15 May 2023 05:33:50 No.7376170 Report

Quoted By:

>>7375975
they're not

Anonymous

Anonymous Mon 15 May 2023 09:50:22 No.7376398 Report

Quoted By:

How long do your gens take anons? Post specs and setting on gen.

Anonymous

View Same Google ImgOps iqdb SauceNAO catbox_cmh0st.png, 1MiB, 1152x768

Anonymous Mon 15 May 2023 15:50:01 No.7376754 Report

Quoted By:

Anonymous

Anonymous Wed 17 May 2023 04:48:21 No.7379747 Report

Quoted By:

Im new to lora training and i need help. I have questions on how to add activation keys for training.

What are steps needed in order to accomplish this. Is there a setting or extensions needed for this procedure.

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 #321