Welcome back everyone!
As a small treat to make up for the downtime, I preset to you, locally generated waifus. Is anyone else running Stable diffusion locally or otherwise? Is there a prompt you'd like me to run for your dream waifu? Are you interested in this technology? Let's talk about it!

Been experimenting with it a bit. Having some interesting results with the image-to-image function using craiyon results as the base image

>>108033
>last pic

So is there an app to download so we can make our own or what, I see Stable diffusion on github but I am not a programmer nerd.

>>108077
>>108077

You can run it on your own machine if you've got the video card for it. 8GB of vram seems to be around the minimum. There's a sticky thread on the stablediffusion reddit that explains how to run it in detail

Holy shit breh
It's amazing how much this stuff is progressing and improving
They future is now and the future is fucking wild

This is incredible. I'll try to get it running on my machine when I get a bit of time to tinker. In the meantime, I'd love to see what it would generate with a round belly/pregnancy-centric prompt

I have a shit PC so I merely ask for someone to run it for me
i'm asking for an AI generated image of a fat Alyx Vance (the girl from half life 2)

im not sure how the prompts work to ask for them, but the goal is a fat/bbw alyx vance

>>108083
Thanks anon

>>108109
It's a bit tricky to replicate characters well, especially if they're kinda obscure (as far as the AI is concerned), but I gave it a shot.

another attempt, a bit bigger

If I may ask, Yennefer from the Witcher...huge boobs...obese...angry.

>>108149

Those are looking pretty good. What kind of prompts are you phrasing to get your level of textured paint and decent look? This is the best result I've made so far.

>>108171
I like to start with craiyon prompts including "artwork by charlie bowater" for decent human shapes.

After that, I use the best craiyon result in stable diffusion's img2img mode, and add in all the prompt fluff that the model seems to like (e.g. intricate details. oil on canvas. official artwork. etc. Can just experiment) as well as changing up the artist styles. Ross Tran, Greg Rutkowski, and Artgerm seem effective, but there's a ton that work.

>>108171

Additionally, I've been doing several progressive iterations. I'll start at relatively high strength (0.5 or so), and then lower it as the pictures get closer to what I want.

I'll use an art program to roughly clean up flaws like those spots on the belly, or other weird anatomy issues if needed, and then use that as the new base to generate from.

(sorry for the double post)

>would need four of my grafix cards to meet the minimum vram requirement
Could I get some bottom-heavy Filia Skullgirls or Cynthia Pokemon with their fat guts spilling over their skirt/pants respectively? Wondering how it'll handle fat bitches with anime faces.

>>108186
Anime features are a bit of a struggle, but I took a shot at it. I've seen other people make some great looking stuff, so it's probably just a matter of finding the right prompts.

Can you do obese chun-li with a fat ass

Planning to set this up locally for myself, what kinds of prompts are you using to get these images? Any tips?

The hell is this shit? It's not worth the waiting time at all

>>108133
>>108140
still impressive work anon, thanks

Can I get a chubby goth girl shorts wide hips huge thighs fishnets

>>108192
>>108207
Fishnets were tricky. The ai kept wanting to turn them into dragon scales.

>>108197
I mentioned some above, but as far as other tips, it's hard to say because this is still super new and people are figuring it out. I generally keep important stuff early in the prompt, and separate each concept with a period or comma. If you're doing multiple iterations, you can sometimes remove a part of a prompt that's well established in the image (like hair color or clothing) already so that it can focus on other details.

Can you do bbw/ssbbw Cz Delta from Overlord with a fat ass?

If Cz Delta won't work, then Kirumi Tojo from Danganronpa also works

I wonder how it would handle immobility.

What about a very fat Valentine from skullgirl eating chocolate

Made this with a combination of the AI and doing some of my own brushwork. I think there's some real potential as a tool for artists here

Requesting Zero Suit Samus with a build about as fat as SSBBW Adeline

>>108083
I feel like I'm in a time loop where I keep typing "pip install <module>" in hopes of getting the damn program to run.

>>108345
https://rentry.org/kretard

This guide offers an easier alternative install method that comes with a great web UI to boot. Godspeed, anon

>>108353
I'll give it a try when I'm off of work. I gave up after python complained that dlib isn't installed then I find out that I need cmake to install dlib, but I need Visual Studio for C++ to make cmake. This is fucking madness.

still trying to get better at the prompts and shit. But Im having fun...

>>108030 (OP)
Fat schoolgirl with extra long skirt.

>>108353
>RuntimeError: Found no NVIDIA driver on your system.

I recently bought a RX 6600 and I don't feel like buying a RTX card. As far as I can tell there's a work around on Linux but so far nothing for Windows. So I give up, thanks anyways anon.

I don't know, something about this tech makes me feel nervous. It's exciting but also kind of scary.

On gawd bruh can I get a hefty tittied obese Anne Hathaway or redhead in a two piece swim suit

This shit triggers my uncanny valley sense really badly. I'll stick to human-made art for now.

>>108584

can I get a chubby dark skin wide hips thick thighs cleavage 🙏

>>108593
Thank you Playa I gotta get into this shit. The quote "we have the technology" is so relevant.

>>108594

technologie!!!!!!

>>108584
Man, I just started messing with this but trying real people seems to create some nightmare fuel faces more often than not. Like they're close but also way off in an Uncanny Valley kind of way

>>108329
Not sure who the reference is, but here's a big Samus anyway.

Just got this running locally a few hours ago. Not having any crazy success with prompt experimentation, but I did accidentally create berry art, so I guess that's something.

>>108699 art kinda looking like one of the characters from that Daft Punk music video

This is like character creation in an RPG run amok with hot ladies. Have some top heavy, regal queens. I got more if y'all want them

Wow. This is really scary from an artist's standpoint. Guess we're gonna be irrelevant very soon. RIP.

>>108840
I've seen a lot of people saying this recently and I really do not think this will happen any time soon. No AI will be able to take and understand detailed instructions at a human level for a very long time, so commissions will definitely maintain their niche in the world - especially since many people pay specific artists to see their specific style, something which an AI will never perfectly replicate. Artists also create more than just standalone pieces! Coherent sequences, drives, comics, visual novels, etc. are all well beyond current AI capabilities. I think (and hope) that artists will be fine.

Txt2img prompt I used: Brunette, curly shoulder-length hair, overweight, beautiful face, detailed face, blue eyes, jumpsuit, full body visible, very fat belly, wide hips, large breasts, looking down on viewer, trending on artstation, glossy

Used GFPGAN and k_dpm_2a in the new Guitard.

For some reason this would never generate below the bust even though I specifically said belly, but I'm happy with how it turned out, especially since I didn't use img2img or touch it up.

Writing "detailed face" or "beautiful face" and specifying eye color combined with GFPGAN seems to help avoid the "uncanny valley" effects. YMMV

It took a while to get something remotely realistic or erotic.

>>108952
Throwing in artist names helps too. Helps make the faces a little more consistent with less wierdness... I think. That's what's helped me

Some experiments asking it to emulate watercolor or Pre-Raphaelite style, this is my new favorite thing (I think the girl is holding fruit in two of them because txt2img misinterpreted "pear shape")

https://vgtim.es/gwzpHvN

>>108799
Absolutely

my pc literally has 16GB of ram, yet for some reason i'm still getting the out of memory error even on the 4GB mode. anyone know any solution?

>>109123
How much VRAM do you have and are you using either the optimized or the Guitard version?

>>108952
I used your prompt and added full body shot and bedroom photoshoot, and was able to get more of a body to generate. I like the results more often with the normalize option disabled. You can get good results by taking something from txt2img into img2img. You can get good stuff by taking stuff that just has the right pose/framing you want and then you can try to use img2img to tune certain aspects of the image. I've put really terrible txt2img images in there and came back with good results.

>>109135
fuck me. checked my specs and it's 4gb vram.

im using the guitard version with the 4gb optimized mode. still doesn't work. i'm done trying and i'll just wait until i can afford a new pc in 4 years, and stick to using my free monthly Dall-e credits.

>>109143
crazy how in only a year my laptop can become so outdated. while searching for the specs on my laptop, i found a laptop that's cheaper than mine when i bought it AND has 12gb of vram.

>>109143
I've been running the optimized version on a 4GB VRAM machine, but it took a lot of trial and error to get working because the GUItard site is misleading. The GUI they have set up isn't actually compatible with the optimized build - even though you can run webui.cmd from the optimized build, it will still attempt to run the unoptimized script and give you the same error.
If you're interested in trying to get stable diffusion running without GUI, the barebones guide (https://rentry.org/SDInstallGuide) is pretty helpful. The optimized version has the same set-up as the unoptimized, but you start prompts with "python optimizedSD/optimized_txt2img.py" instead of "python scripts/txt2img.py."

I got asshole titties somehow

>>109070
More noble babes

>>109070
Started messing with adding celebrity faces in the prompts plus artist names. Nigella Lawson works good in my book

I generally just run a bunch of text to image prompts, then once I get a good one from there, tweak it a bunch image to image. Thats what a lot of these are.

Full body shots seem to result in messed up faces too. Not sure how to get better results with those.

Damn son, even an RTX 3070 TI isn't powerful enough for this. If the input image is too big then I get CUDA errors, and if the image is too small then the AI doesn't know what to do because it wasn't trained on ants. Anyone have an idea to solve this with an 8GB VRAM limit?

Can it generate a photorealistic image from a drawing as the seed? Suppose you started it off with your fav artist...

>>109198
keep getting this

"import torch
ModuleNotFoundError: No module named 'torch'"

>>108965

Out of curiosity, what artist name(s) are you using for the prompts of these lovely regal ladies?

Hi all, some of my attempts with ai generated images. If you want see more
https://www.deviantart.com/jaan9

Hi all, some of my attempts with ai generated images. If you want see more
https://www.deviantart.com/jaan9

>>109256
Usually at least Greg Rutkowski, Artgerm and Alphonse Mucha. Those 3 yield decent results. On alot of them I've also added in Mark Arian and Kenneth Rocafort and Alex Ross too. Past a certain point with the prompts, I think you add too much and its hard to tell what effect its having.

>>109245
I've been working with 8GB so far, but I have to use the optimized scripts. The standard ones run out of VRAM.

>>109276
Would you be open to writing some of the initial prompts you used here or in the DA descriptions? Especially for the last two, I usually struggle to get the AI to draw different camera angles or poses.

>>109245
You may need to manually resize the image and force the output image to always be 512 x 512. I had improvements in both efficiency and accuracy once I started manually resizing source images in the Linux version of MS Paint.

>>109248
Weird, the environment.yaml step should've covered that. Try "pip install torch" after "conda activate ldm" and it should (hopefully) work.

>>109355
i just went on a whole ass journey, trying to run this fucking thing, but now i'm at a roadblock. conda is telling me that i need module "cv2", and due to the python file i'm reading, im guessing it will ask for module "time" as well. problem is? pip install isn't working. and i'm tearing my hair out because i've never done anything remotely as technical as this.

Here's some foolin around i did.

>>109396
some more for your veiwing pleasure.

a few more. Started masking/blending results together for better faces/hands.

Was trying to get some game characters like Yen or Triss from The Witcher and seems like the AI model just doesn't have much to draw from for them. Not great results

Mixing famous faces is interesting though- Margot Robbie + Nigella + Anne Hathaway. Or getting a good prompt and then just adding a word or 2 at the start to change things up- so instead of "Curvaceous obese regal queen Nigella Lawson" add in "Curvaceous obese vampire regal queen Nigella Lawson"

Pretty happy with what you can get out of this

Started with
obese naked ssbbw wonder woman,huge fat belly,Gal Gadot,by Greg Rutkowski,by Artgerm,by Kenneth Rocafort,by Alex Ross
Steps: 16, Sampler: k_euler_a, CFG scale: 12, Seed: 1319421247, GFPGAN

Then used image to image settings:
obese naked ssbbw wonder woman,chubby face,huge fat belly,pregnant,fat,huge tits,Gal Gadot,by Greg Rutkowski,by Artgerm,by Kenneth Rocafort,by Alex Ross
Steps: 50, Sampler: k_euler, CFG scale: 9, Seed: 2570754029, Denoising strength: 0.75, GFPGAN

Through several pushes of better images to get this result

What is everyone else trying for settings?

>>109459
I typed in 'beautiful fat woman in the style of mark aria', and set the program to generate 250 pics.
when i woke up i had about 198 really good portraits of fat ladies. the facial correction program works really well too. it can take a bad photo and make it great.

>>109474
if anyone is interested, pick a number from 0-197 and i'll post the picture of your new supersized mail-order bride.

>>109475
151

>>109475
69, obviously.

6

>>109459

Made a little gif animation showing my craiyon >> img2img process. I didn't spend as much time on it as some others, but it gets the gist across.
I used a new seed for each step.

>>109475
100

>>109476
She enjoy's long dinner dates and short walks on the beach.

>>109484
I made a whoopsie and skipped or deleted 65-70 so i gave you fat brittney spears instead.

>>109485
She looks like the type of girl who would enjoy dancing if she wasn't so out of shape.

>>109488
It's interesting how these turned out. i can see features of famous actresses in some but i used no names in the prompt.

also, this one looks like she's going to double the size of your grocery bill.

>>109487
I didn't know you could crop and paste like that. thanks!

>>109507
>>109509
>>109512
>>109513
if you see these and like them i have ~190 more. pick a number between 0-197. I'll be on again later tonight to post the results.

Hit me up with some prompts you want to see. Give me a celebrity or celebrities, preferred clothing/outfit, maybe a certain body part to emphasize/focus on.

My 3080Ti awaits your command

You know, I think some of my uneasiness regarding this tool comes from the thought of how easy it will be in the future for someone to use gallery-dl to rip a handful of artist galleries from Twitter/DevinantArt/Pixiv, using them as image sets to train the AI on, and then using the images it generates as designs to be printed on mugs, posters, t-shirts, and more. And that's just one scenario, y'know? Outside of art, who knows what the fuck this thing is going to be used for. How are we supposed to stop unscrupulous political groups from making themselves appear more legitimate with graphics that look almost professional? What if someone with a petty grudge uses it to flood image boards like this one with a bunch of uncanny gore or CP? What kind of havoc are closeted discord groups going to wreak once they automate all of the above with bots?

I get that the internet was already a horrible place 24/7, but I think the last thing it needed was something to exasperate those qualities even more.

>>109538

I think the concerns for artists are legitimate. At the rate the tech is advancing, I expect it will cause some tumult in the "artwork on demand" category of jobs. Artists who make art as their passion won't go anywhere, but it might get a lot harder to make a career out of it.

For political stuff, photoshop has long taught people to not believe everything they see without question. As people becoming more broadly aware of AI's capabilities, I'd guess trust in photographs and video online will continue to decline.

As for things like gore and CP, people already can flood boards with real pictures of that stuff if they're so inclined. I don't think the ability to generate fake versions really changes that very much.

>>109541
Aw, yeah, I guess you're right. IDK I'm kind of getting whiplash at the rate this is advancing. There's like a Photoshop plug-in being worked on right now.

>>108030 (OP)
Wow, Stable Diffusion is really good at making pictures of Wonder Woman, like more then any other fictional character I've tried.

>>109561
Like, here's my best attempts at making Chun-Li.

>>109562
>>109561
And here's Psylocke.

>>109561
it probably helps that wonder woman has is probably the most used superheroin when it comes to sexy picsets. both fat and skinny models alike.

>>109534
Shit Hayley Williams in red lingerie massive Breast and pushing about 350 pounds

>>109590
Quick attempt just text to image, just ignore the mutant hands:

>>109590
Some more decent ones from the batch

>>109487

Your process is top tier and I too felt compelled to make a hooded beauty as well. Bravo for the walkthrough!

>>109594
Bruh thanks, yeah this stuff really is wild and a cause for worry but it's awesome at the same time. Once it's turned into an app people can just commonly use its over with for real

>>109487
What img2img did you use?

>>109598
Yall making such great results sheeesh

>>109635

I'm using the optimized img2img python script via miniconda (basically just followed the guides in the stable diffusion subreddit).

>>109652
Wow i jave no idea what you talking about but god fucking damn it those are some amazing fucking results.

I tries to use art breeder. Could you try aomething with this? Not really on a PC currently(also actually thought there was a site for this no judging please i'm alreary aware how dumb i am lol)

If it isn't a trouble coud you do your thing if you think of these is worth a try?

>>109654
Ok the first pic i toom from dall-e and it is small for some reason

>>109534
Alexandra Daddario please. Swimsuit or cowgirl attire? Ssbbw with a fat ass

>>109654

>>109654
I didn't really know what any of it was either, I just kinda followed along until it started working.

The small size is normal from craiyon (dalle-mini), but it's not really a problem for img2img.

Can somebody try generating something more 'toon styled? Or just stylised in general. I would do it but my RAM is lamentable

>>109672

I tried running the same seed image and base prompt through with a few different styles. Don Bluth/Disney, WLOP/Studio Ghibli, and Dreamworks/Pixar respectively.

>>109686
Not bad! I may be made redundant soon!
Although, I would say these are slightly lower quality compared to the realistic/painterly ones. There's still some room for improvement for v.2

>>109698
Part of the quality may just be the prompts I'm using. I've had time to refine the exact series of words for the less cartoony style and see what effects it has. It might be possible to get much higher quality stylized results, I just don't know how yet.

>>109534
Rosario Dawson and/or Zazie Beetz please. No strong feelings about the rest, but if it's helpful, maybe swimsuit or lingerie?

>>109487
Could you run this craiyon image through your process? Thank you.

>>109704
I ran it through an upscaling AI as well to get a somewhat bigger result.

>>109709
OH WOW! This looks fantastic! Although, i can't help but wonder if the AI can make them fatter...

>>109711
It definitely can, I was just trying to maintain the general shape of the original picture through iterations.

Heck, you can even make use of the ol' handy sphere tool between iterations, and just have the AI correct the distorted look while keeping the new size pretty effectively.

>>109704
I am but a lowly peasant sire

>>109715
Really? Can you show me how that works?

>>109719
Basically this.
I blended the head/hands from the original back over the new body for the final image.

>>109722
Man, if i had access to a powerful enough PC I'd definitely do this. This tech is incredible.

>>109709
Would it be possible to upscale this image with the same subjects? I.e. have them have similar faces?

>>109725

One of the current struggles with the current AI is consistency. Making a cool character once is easy. Making the same character again is borderline impossible.

I can try to make a picture with that image, but they're very unlikely to look like the same two people without hours of generating faces.

>>109726
Ah ok, i see. If it's alright with you, I'd still like to see it upscaled.

>>109722
Hey the previous anon here.

I was gonna respond that the previous one looked great but goddamn you fucking made it better!

Btw do you happen to be iven for aomething like commissions? I was wondering if in the future i could maybe ready a batch of images for you to do your thing on your side as i have no idea how to do it and i'm sure i don't have the hardware

>>109725
Gotta do some other stuff today, so going to be giving my poor video card a break for a while.

>>109730
I don't know how I feel about that prospect. Aside from using my computing power, I'm not really contributing that much to making these really. Just the basic prompt that I already shared, and some very minor touchup in photoshop.

>>109733
Yeah that is the point. I don't have the computational power so lemme at least pay you back the energy cost? Maybe for your time and patience too?

I really don't know how much would be good tho. I don't know how much the energy cost would be or how long one of these take.

>>109733
not the guy who asked originally, but i've got a HUGE amount of craiyon images. if you'd be open to it, i could send over a zip or something, and you can knock them out whenever you feel like it. zero rush, just something you could have for when you're bored and don't know what to do with the ai.

>>109652
Damn, those are some nice legs- did you get those from any kind of text to image prompt or just from craiyon to img2img?

That's the one thing I find a huge crapshoot in text to image, good legs/butts are hard to come by with any consistency.

>>109660
These are just text to image. Going to try and touch up some image to image ones later- alot of the full body shot ones are horrible abominations once you get to the legs.

Some Cowgirl and bikini Daddario

And for people fearing the death of art from AI, it shits out mutant hotdog Alexandra Daddario just as easily as it does good images

>>109781
No dude you don't get it thisbis a new fleshlight design. AI is still scary it even gives us what we disn't ask for

>>109475
175?

>>109779
>>109781
Good stuff and holy shit this is hilarious>>109786

>>108588
YOUR WHAT VALLEY SENSE?????

>>109802
https://en.wikipedia.org/wiki/Uncanny_valley

I am ricky retardo, I have a 1660ti and I am running into a memory allocation error, is my dingy card not good enough?

>>109802
shut the fuck up. you're not funny.

Using the words "action photo" and "detailed face" seems to produce decent waist-up shots. Even if there is no action in the photo.

Adding celebrity or character names also seems to help make a more normal looking face. Then run it through a few passes of GFPGAN to fix any weirdness with the eyes and mouth.

more trying

>>109534
Ariana Grande, thick and pear shaped and/or with a big bubble booty
Outfit should be something like these little shorts she wears or whatever they're supposed to be called

Oops, wrongly sent this to bbwalt

>>109701
For whatever reason, Rosario Dawson and Zazie Beetz look pretty terrible in this. At least just doing text to image prompts they're like goofy bad caricatures. Seems to be the case with many celebrity faces though- the AI makes them like caricatures to some extent, exaggerating facial features.

>>109960
Do you remember the prompt you used for the 3rd one? It looks incredible

>>109960
2nd and 3rd look incredible

>>110081
Hard to get txt2img to generate good full body shots without the face looking like trash; these are all just tx2img

>>110081
Other issue I get is sometimes you get good body shots (last 2 here) but the face is cropped out. Anyone know a good way to try and get the face included? Img2img doesn't seem to work to include the face when I've tried.

>>110081
These ones I tried cropping the face out of some full body images that weren't great and running the face under img2img then pasting it back in and running img2img again. Turned out ok, just ignore the mutant hands.

Try Jade Harley from Homestuck

I suppose I have 2 different requests, for 2 different things

1. a massively obese Mistral from Metal Gear Rising
2. for the guy who is trying the realistic shit with actresses, could you try making a fat Elizabeth Debicki with it?

>>110111

That one I actually started as two craiyon images that I individually upgraded slightly with img2img, then spliced together, before running back through img2img again to tie them together.

As for the prompt, to my best recollection it was probably something like "two voluptuous women. chubby. big hips. intimate. beautiful highly detailed digital illustration. artwork by ross tran, artgerm, greg rutkowski"

I might have included other details for individual generations to try to steer it, like hair color or clothing.

is there an uncensored version of stable diffusion online? as I dont have a pc powerful enough to run it

Could someone run this craiyon image through img2img and upscale it?

>>110196

whats the minimum specs needed for this ai stuff?

>>110200
I've seen people say as low as a 4GB VRAM video card, but in my experience, even an 8GB one requires some compromise and using specific optimized scripts to function.

>>110201
oof, I have a 4gb VRAM card

what would need to be changed for it to run?

Can someone upscale these craiyon images?

im stuck on the tutorial
im on :

Step 9: If you have any existing folders in /src, delete them before the next step

Step 10: Run the following command: "conda env create -f environment.yaml" and wait
(Make sure you are in the waifu-diffusion-main folder)

Step 11: Download the GFPGAN pre-trained model and place it in src/gfpgan/experiments/pretrained_models/

but uhhhh, I dont know what or where I can even find this /src folder to begin with

>>110220
The tutorial isn't quite the same as the one I followed, but I believe the src folder should be inside the waifu-diffusion-main folder it mentions earlier in the steps.

>>110224
it is not

>>110081
I guess someone beat me to it but okay

can anyone run some face fixing on these images?

they're meant to be moondragon from marvel comics, I know the last one isnt exactly BBW but its one I generated that I really like

I generated these over a website which is why im asking for face fixing rather than doing it myself

>>110148
Debicki

>>108030 (OP)
>>110069

The thing I've found to make the face work most of the time, though not all, is if you're using a character played by an actor/actress, make sure to put something like 'x as y' or 'x played by y' or something similar. Having both names seems to make the ai cross reference the person and that generally creates better faces, imo.

>>109487

one flaw/problem I'm having, and maybe it's just me, is that I can't get the img2img to recognize larger figures/breasts when I put them back through again. I can generate images, but when I edit them in photoshop to then run through again, it doesn't seem to want to adhere to the figure I give them, I guess? It's a bit frustrating.

Have you found that the ai prefers certain things over others for this?

>>110280
I've found that tinkering with the strength parameter can help with that. If the strength is set high, it might take more liberties with low detail parts of the original image and just do whatever it wants.

Lowering the strength for a few iterations while emphasizing the parts you want it to keep in the prompt can guide it a bit more. Once it's accepted something is a particular body part at lower strength, you might be able to turn it up for cleaner results.

>>110133
>>110134
>>110139
>>110233
Bruh y'all some real ones 🙏 was not expecting this many

>>109534
Yumi Nu in a tiny bikini top and a short skirt large breast large belly. Hope this will work when you give it a woman who's already chubby. (Use Miranda Cosgrove if Yumi Nu doesn't work) As a side note, I wonder how this thing can handle requests for breast veins, stretch marks, tears in clothing and more dynamic poses

Downloaded it locally and found that Christina Hendricks was one of the people I had the best results with

I've posted more here: https://www.deviantart.com/aigeneratedladies

could anyone try img2img with this?

its mistral from metal gear rising so maybe try prompts like
"mistral metal gear rising, extremely fat, bbw, realistic, big belly" etc etc

>>110371
what im asking is trying to get this in a realistic style with img2img

Have people had luck converting photographs (e.g. of a curvage model) into digital paintings? I always get really weird results even when I scale to 512 x 512 for both.

>>110408
I haven't had much difficulty with that. Usually can get some decent results even on a first iteration.

By weird results, do you mean like slightly weird faces, or entire body misshapen issues? If the latter, the strength may be set too high.

anyone have a good installation guide?

the one in the thread doesnt seem to work for me

Just use this one, the tutorial take a couple minutes but explains step for step what to do https://github.com/hlky/stable-diffusion/wiki/Installation#windows---step-by-step-installation-guide

>>110442
is this one uncensored or is there a tutorial to disable the censor for it?

>>110306
Tried Yumi Nu but she looked like some generic asian lady, not really like her in any way. So have some Miranda Cosgrove. She came out a little better. Just txt2img.

>>110306

>>110306

>>110306
Last batch

Some more

>>110479
thanks, I'm in the process of getting this thing up and running on my computer, so hopefully I can do some thread requests

>>110459 There is no filter in that version as far as i can tell from what it puts out.

>>110103
Appreciate the effort. If anyone is able to get it working for some black/brown women that'd be awesome

I am awestruck.
This is the future we have chosen.
I'm okay with this.

>>109487
I saw this gif over on 4chan and thought I would try my own picture following the steps you lined out. I'd love to do more AI pics, but I don't have enough VRAM to use SD properly. I'm only able to make pictures by running SD utilizing the CPU.

Paintings were made with Dalle2 and image editors to composite the images and coax dalle into making huge sizes.

Pictures were done with local Stable Diffusion. The topless one was touched up with img2img

More: https://www.deviantart.com/robojuxtaterrestrial

>>110881
What sort of prompts did you use to get the Dall-e 2 ones? I'm kind of wary of experimenting on my own with it.

>>109534
Bbw Jennette McCurdy wearing a bikini or bodycon dress that highlights her body fat

>>109534
Nicki minaj, emphasize on the whole body.

>>110881

Experimenting myself, I've never managed to get SD to read large stomachs right, or any kind of major expansion. It doesn't know how to read giant body parts, and seeing as textual inversion of any major datasets requires upwards of 40gb vram, I'm not sure we ever will.

Image to Image doesn't really help either; it can't read large stomachs or breasts. So I'm guessing what you did was mesh them in photoshop and not actually use SD to make the initial images, because any stomach larger than a standard pregnancy isn't really recognized by the SD model.

I'm not actually certain it knows what 'fat' means either, because it really seems to struggle with anything even slightly off the standard generation path. Hopefully one day soon people figure out how to actually make it learn things without needing five gpus to train it.

tried using that new image extension thing dall-e 2 introduced.

Trying the Waifu diffusion finetune from reddit. It definitely has an easier time with faces and forms, but also just randomly adds breasts to things when not trying to generate women sometimes.

>>111372
> Waifu diffusion finetune from reddit
reddit jannies already removed the link, lol.

>>111396
reddit mods exits only to destroy any fun ppl may have lol

>>111396
I double checked and the one I used at least is still up.
https://www.reddit.com/r/StableDiffusion/comments/x8y1u3/waifudiffusion_v12_a_sd_14_model_finetuned_on_56k/

i remember reading that there was a way to disable the nsfw filter on colab versions of SD, but i completely forgot how. anyone know anything about this?

>>111413

Question. Can you use multiple models at once? Meaning, can you use waifudiffusion and say, 1.4 at the same time or do you have to choose one or the other?

>>111464
Not to my knowledge. I think waifu diffusion is just a slightly modified version of the normal 1.4 model anyways, so I'm not sure it'd be useful to combine them even if you could.

Don't know how to use it so i Guess Try Megumin with some Thicc Thighs

>>111413
Ah, thats not the result google gave me,
https://www.reddit.com/r/MachineLearning/comments/x9d4tc/p_waifudiffusion_a_stable_diffusion_model/

>>111489

>>111690
Any chance of a bigger, more BBW-ish Megumin?

>>111703

>>111711
About about megumin with a massive ass? Back view if you can

>>111711
That's really cute! Thank you.

What prompts are you using/how many times are you all running the waifu diffusion for? All I'm getting is distorted blobs

For you animals

here's another

We can use any model for this? Could I use plump princess or azismiss?

>>111188
>>111357

Dalle will not give the huuuuge bellies. I used image editors to make the art ones. But not like drawing.

The basic dalle technique is to get it to make a good face. Download, then scale it down, then re upload it and have dalle fill in the details. Repeat.

The problem is that the more times you scale it down, the less of the really good original image there is for it to piggy back off of. You end up with the 'drawing' technique messing up.

the one with the girl laying on her belly was made with a similar technique, but instead of scaling down, i just moved half the picture off the canvass and re uploaded , then just put the resulting image back together in an image editor (basically dalle2's out painting feature now.

The way I did huge bellies with dalle is by tricking it. It takes a lot of luck and a lot of regenerations, but if you get the right colors in the right place, using the technique i outlined, you can do stuff like that.

The trick i used for the chick laying on her belly is the whole image was actually made with the girl laying on a yoga ball. Then once i had the rest of the anatomy good enough, i put it in an image editor, and pained the ball very roughly with skin colors. Highlights. Shadows. Really ugly basic stuff. Then one small section at a time (because it's a big image compared to dalle's 1024X1024) I had dalle regenerate parts of it. Like the edges of the belly. then the internal shading. Again that takes a ton of regenerations and luck and careful prompting. probably took 200 credits for that one.

the two phots are made with SD not dalle and are basically just made with careful prompting. The one with the chick laying on the bed though was touched up with img2img in SD for the nipples and hands.

>>112175
Holy fuck, I can barely tell that the one on the left came from an AI instead of Pixiv. The only tell is that the eyes, fingers, and shoulders are a little fucky and (maybe) the lack of a navel, but it nailed the rest of it, complete with extraneous swirly bullshit and hanging foodstuffs.

For the love of God, G*d, Allah, YHVH and the rest of them, please post some more along those lines if you're able.

Hey can someone do like Mikasa from Attack on titan with an absolute dump truck of an ass a back view would be good

>>108323

I find the faces in this thread quite bad, I think if an artist was good at faces and could redo/combine their face drawing skills with ai everything else it could really be a) good and b) a workflow thing.

>>108799

these faces are amazing, my prior message doesn't apply to these

>>109594

she looks familiar (face) what celebrity/actress is this based on?

>>112250

>>112175
upscaled and edited a bit

I've had little luck with generating images, sometimes taking 100's of images to get anything useful.

I then run the ones that aren't absolute nightmare fuel through gfpgan but even that doesn't always fix the faces

Out of the maybe 20% of images that I deem "okay" from stable diffusion alone, roughly only 10-30% of those come out okay after gfpgan

I can't help but wonder, am I doing something wrong? I run 60 steps per image, include keywords like "detailed face", "artstation", "sharp focus", and other prompts that have done well on Lexica.

Attached image is the very best of all my goes and the only one I consider to be very good, but even it isn't perfect. I had to fix the hair myself and remove the janky background in photoshop.

>>112329
Thank you my man

>>112337
It's looking really nice, I'm looking fowrward to see more pic in this style

duperboyprime has been making AI art and combining it with photoshop

Could anyone do Mai Sakurajima from Aobuta please?

>>112502
Yup, that's exactly what I said would end up bringing us a new wave of art not too long ago on the previous AI art post.

>>112231
thanks, here's some more. Not all of them are home runs, but it's nice when the ai generates a rare great one. I haven't had great luck with having the WD model generate good fat pics just from text, but I've had promising results with img2img. Maybe the txt2img just requires more patience.
>>112337
looks good! I think a lot of the current models' pics would look even better with some manual touching up
>>112412
the artists I used were from some prompts on /g/. that one was "(Albert Lynch), Carne Griffiths, Franz Xaver Winterhalter, Lilia Alvarado, Sophie Anderson"

>>112329
>>112385
Decided I'd make Mikasa's ass larger. Getting a good result with img2img at this size was tricky, so I used image tweaking with a low denoising strength, but I like how it turned out.

Tried few things in Craiyon

And also quick Edit thru DALL-E

>>112535
How do you do something like this?

>>112535
This photo is fantastic man looks like it was drawn. Hope you can do more characters like this like megumin or Asuka

>>112627
I prompted for a pose that looked good then edited it in photoshop for the size and details I wanted. Then used a low denoising value to keep the details and shape intact. Also here is the upscale I forgot to make.

Can we get a really fat Abigail Shapiro and Krysten Ritter

how do you bypass this bullshit

>>112813
Nice work what prompts are you using for these reults?

Upgrading from Craiyon to Stable Diffusion is wack. The detail improves but the girls get thinner. Many of the Craiyon prompts used to embiggen just don't work here. I can't even get Gabourey Sidibe to look as big as she used to be in real life! We need someone to train SD on better images and make that available.

Oh, and the porn one at https://pornpen.ai/ has a lousy selection, basically the "fat" button could mean anything from 175 pounds to 600, usually the former.

>>113036
From What I Remember:
(name),((obese body)),bbw,cleavage,(pretty face),Michael Garmash,Justin Gerard,Greg Rutkowski,Alex Ross

Negative prompt: naked

CFG:12

>>113121

One thing that's worked for me is that it seems to give different kinds of shapes depending on the prompt. So using waifudiffusion as a model, 'belly expansion' gives a different shaped belly than 'pregnant' does. Though ironically, 'belly expansion' gives more pictures of large breasts than 'breast expansion' does.

I've also found that using 'chubby' creates very large shaped women.

One other thing I would mention just from generating lots of images: using multiple artists really confuses the model. what I usually do is pick one and then place it like fourth or fifth in the line, so it'll use it.

>>112576

Okay, 100% I have no idea how you did this. I guess using img2img and masking? Because in my experience none of the models that have been released know how to actually do anime in the slightest.

>>113020
easy, use a host that isn't bullshit. This AI is open-source and can be edited and downloaded/shared by anyone.

I personally use the google colab as it allows for NSFW, has support for img2img, and has TONS of options. Outputs appear in your Google Drive.

If you have a powerful enough computer (I sure as hell don't) then host it locally, 8gigs of VRAM are recommended with 6gigs being the minimum.

Here's the colab link: https://colab.research.google.com/drive/1jUwJ0owjigpG-9m6AI_wEStwimisUE17#scrollTo=Ucr5_i21xSjv

if anyone has better hosts that they use please do share, I can't figure out img2img lol

>>111372
Do you have to run waifudiffusion through the python command line or is there a GUI for it?

>>113290
I'm using the GUI with it. I just swapped in the waifu model and then renamed it so it'd use it instead of the base model.

Been messing around with trying to make some bigger women. Also had a go with that Abigail Shapiro from a few posts back.

>>113409
And a few more of my recent ones, having some good luck tonight.

>>113410
These are pretty good!
What's the prompt for the girl with the blue shirt at the computer?

how big could you push the AI?

probably not saxxon but could you do something like satsurou sizes?

>>113410
The Chinese themed one is nearly perfect the only thing throwing me off is her hand. Anyone willing to do a quick edit?

I've been getting it to work and still learning, however when using craiyon as the base image make, then switching stable what do I input for the prompts to fix faves? As well as mask and inpaint image? more Craiyon generated images?

>>113579
small phone sucks, fix faces and in stable diffusion

>>113579
When you get an output that's okay overall. Crop the face and a bit of the shoulders and use img2img at about. .6 or .7 strength. Add things like "cute face, bedroom eyes, or detailed eyes" to the prompt so the AI knows what's up. If you want a face that looks like a particular person you'll want to turn down the strength of your input image so the AI has wiggle room.
Im not at the PC but I can post examples soon.

>>113583
thanks trying it now. Just to let you know I'm getting good results put of craiyon just with some rough anatomy and horrid faces. So I'm trying to refine the results installed diffusion

>>113583
also any tips for how Inpainting works would help too

>>113503
damn this is like almost fap worthy

>>113461
Sorry brother, I've only just gotten into the habit of saving my prompts properly. Also on the high trying to chase the last picture with the red head again, holy fuck.

>>113547
now u don't needa worry bout the hand anymore homie

>>113754
forgive this monstrosity, I offer shrine maidens in recompense

Alright, stayed up far too late doing horny stuff, so have these.

And if anyone was interested, here's some iterations I did with img2img.

Anyone got a tutorial or prompt guidance? I'm not sure if I'm using the google collab correctly and just need a bit of explaining please

Playing around with some stuff, I'm still trying to figure out how to improve the faces.

>>113797
> stayed up far too late

this shit is dangerous

I gave it a try today, these are my results so far

>>113583

I think you need a bit more explanation for that. If you're cropping things, you're going to need to also resize them to 512x512 right? And with such a high denoising, it's going to make it nearly impossible to place over. One of the big weaknesses of the method you're describing is that you simply do not get good results; and the reason for that is that the more you try to fix things in img2img the less the things are going to match up. At best what you end up with is a bunch of grids.

That's a huge flaw in the model that can't really be fixed, due to the fact that it's trained on 512x512. Because of how denoising works, you can't just generate larger images to make them better. And at 512x512, details are pretty lacking.

One thing that I would like to see is whether or not anyone can get textual inversion to work, and either mimic a certain body size or mimic an artist style. At present, I've had no luck with it, because I keep running up against limits of the model with understanding what an object is or what a style is.

>>114417
I haven't needed to resize anything for inputs. It seems to accept whatever size, and just resizes for the output.

When you paste the new face over the old one, you'll end up with some lines where the new generation starts. If you use an art program with layers and soft edge brushes or masking, you can clean that up pretty easily though.

>>114435
Would you happen to have a link for a good FAQ on how to use img2img with prompts?

>>114435

You're going to need a lot more detail in your explanation, because based on what you've said, it goes against how the program itself works. Are you just cropping and then layering in photoshop? Because you should logically need to resize things if oyu want details.

>>114460

Not sure I believe any of these were AI generated without the prompt you used.

>>114568
I don't have a FAQ unfortunately. I've just been learning via trial and error personally. For img2img prompts, I generally use the same sort of prompt I would for txt2img, with small adjustments based on what I'm trying to change. For example, if I'm cropping a face to re-generate new ones, I might change the artist to Artgerm, or another artist who has a focus on face close-ups in the model.

>>114574
I was previously using the NMKD GUI, and have now switched over to AUTOMATIC1111's web GUI. In both cases, the img2img function hasn't cared what size of image I use for the input. It'll just resize it down or up, and squish or crop it based on settings if it doesn't match the output settings.

For cropping and layering, I am using photoshop. The areas that I crop out and use for a focused generation (such as the face) get upscaled in the process (i.e. from a 137x137 crop to 512x512 output from the AI). I then have to shrink them back down to fit the original image, yeah, but they still get a nice quality/coherency boost by being generated separately.

I hope that's more clear.

After jumping through a gorillian hoops to get this thing working on my 1660, I decided to throw one of my drawings into the ringer.

I for one welcome our new robot overlords.

>>114575
I'm pretty sure this person is just dumping a bunch of random images from their hard drive, expecting other people to make generations for them.

made with SD1.4 by chaining a ton of img2img with essentially Jessica Simpson Tinker Bell
though hakurei's waifu diffusion seems to work best when going for common traits (drawn, secretary suit, big but not massive breasts)
cleaning up some almost perfect ones in Krita before diffusing again
lower strength when you want to hone in if you think the ai has a good idea, raise strength when you reach a "local minimum" and want more variance

>>114587
Entirely possible! The reason I said that I didn't they were AI generated is that, having spent hours trying to get the AI to mimic styles, and even spending time using a colab to try and train the model using textual inversion, it's very, very hard to get it to actually mimic a style. Beyond that, it's really hard to get it to have that kind of clear cut drawing. Here, I'll post some of my own examples as evidence of what I mean.

All of these were done in a supposedly 'comic' style using various comic artists. You'll notice that they share similarities, and that's because the AI model doesn't mimic artists styles well at all. Even famous artists from history, it doesn't mimic well; that's just a limitation of the model at present. Beyond that, the AI doesn't generate hands, specifically fingers, well, nor does it generate expressions well.

What I mean by that is that if you look through the laionB dataset you'll see that if you put in something like 'happy face' or 'happy expression' you'll get a bunch of non-human stuff. Even if you use the right keywords, faces generally come out distorted without a mix of luck and refinement; I've had to generate probably upwards of fifty thousand images to get around a thousand decent ones. To put it another way, in a batch of 1000 images, around 30-50 are actually passible.

Going beyond that, there's the matter of handling multiple subjects. The AI does not like that. It does not like it at all. It can barely figure out that humans have two arms, and frequently generates them with three or more.

Then there's the size. Now, the model that people use with stable diffusion is trained on 512x512. That doesn't mean you can't use an upscaler. But there are artifacts when you do that, and it shows.

All of these are, imo, limitations that may or may not be improved in the 1.5 release.

I will also say that the AI model does not do fatness well! Ironically, It understands how to model the pregnant body well enough, but not really how to generate fatness in a way that doesn't seem distorted, at least in my experience?

Then again, I've never managed to get img2img to work effectively. I only really use text2img, and it's often a crapshoot what you'll get out of it.

>Inspired by some SD body horror
>don't know where else this should go.


Ouch...oh, fuck!
Sara's hips, actually her whole pelvis hurt so bad. They sad it would, but this was way, way more than expected. She stirred in the hospital bed slowly, moving ever so gingerly to avoid the sharp pain in her lower half. No use.
Sara had just woken up in recovery from her pelvic widening procedure. She had long admired the girls she followed on Instagram with the cute little thigh gaps but with her build this was her only option.
It had been a gift to herself for her 26th birthday last month when when she went to the consultation and put the money down for the procedure, well actually, she charged it. This was starting to feel like a big mistake.

Sara tried to roll on her side. Ugh... What the fuck? Owww! Her eyes welling up a little at the pain. She tried again, Ugh! her lower half hardly budged in the bed. What the hell is going on? She said as she experienced a really strange feeling other than the pain. It was like she was trying to rollover a big flat rock, but the big flat rock was her lower half!
She shimmied to the edge of the bed and swung her legs over the side of the bed. This was going to hurt. One toe cautiously made contact with the floor. Ahhhh-ah-ah! Son of a bitch! pain shot up her leg instantly. Something didn't feel right but Sara was scared to look under the blanket covering her legs yet.

Sara saw there was a mirror on the wall next to the bathroom door. If she could get over there and access the situation maybe she could put her mind at ease. All this fear was just a side effect of the anesthesia wearing off, she told herself.

She lowered the other leg to the floor. Oooh! Shit!! Owwww! It felt weird but she pushed herself forward off the bed almost loosing her balance for a sec. Sara didn't want to look down, not till she got over in front of the mirror. It was only 2m away.

She wrapped the blanket around her and began to shuffle toward the mirror. Her eyes squinting in pain.

As Sara approached the mirror the general shape of the person she saw looking back looked like something out of a fun house mirror.

Startled, Sara let go of the blanket.
Her new figure was revealed as the blanket dropped to the floor.
Sara got a lump in her throat and her jaw dropped.
"What in the fuck did they do to me!?" she mouthed, though her words made no sound.

During the consultation she had decided on a cute 7cm gap. Just the perfect amount that looks great in swim suits and yoga pants.
But what stared back at her in the mirror was something else entirely.

Standing before her was a woman that looked like Sara from the waist up, wearing a white post-surgical compression corset. But the woman's lower half was hard to put into words that did it justice.

Between two otherwise normal looking thighs was not the 7cm gap Sara had picked out. There was a SEVEN INCH GAP!!! Her legs were almost 18cm apart NOW!
She was expecting to go from 37inch hips to 42.5inches.
Instead of adding about 2.75 inches to the front of her pelvis at her pubic symphysis and another 2.75 split between her two sacroiliac joints she had a total of an extra 14inches added to her hips.

50 inch hips, 50! inch hips! What was she going to do now? That combined with her 26 inch waist and her above average 32F chest made her look like a caricature of a slut, not the cute, toned, gym girl she had been working for the last 3 years to become.

As Sara caught her breath she started to started to cry.
Fwwwwppppt...FFFWWWWPPPTTT.....FWWP..FWWWWPPP...FWWWWTTT. Things just got way worse Sara realized.

As her diaphragm spasmed while she bawled her eyes out, she came to the realization that her legs were so far apart now that even with her knees touching that if she took any sharp sudden breaths, air would get sucked into her now much roomier pelvic region, then pushed back out and make those noises as it did so.

Sara also could feel heaviness that she assumed was her organs having shifted lower due to the now excessive room in her pelvis and nothing much to hold them up in place. In fact the corset even looked a little loose at the waist. Normally Sara would have been thrilled with such a change, but all that probably meant was that her waist had actually shrunk. Further expanding the ratio of her waist to hip measurements. "What was the ratio of 23 to 50 plus?" Sara wondered for a split second.

With her arms hanging at her sides behind her hips she realized they were way, way wider than her shoulders now. She couldn't even swing her arms without bumping into them
Sara stopped crying for a second and brought her hands around the front and down to her crotch. She cringed as she saw that she could put them side by side on her mound with room to spare. She felt as ridiculous as she looked.

>>114706

Hi

Girl with, big boobs, penis, red hair

>>114694
Interesting take, I partially disagree. In my experience the AI seemed to have a surprisingly solid grasp on fat distribution. Might just be because I'm a brilliant artist who gave it the best human-made art to work with, as shown in my post from yesterday: >>114586

Banter aside: I'm super curious what your prompts looked like, I was putting in things like "Voluptuous, plus-sized woman with massive thighs and a big soft belly" avoiding the word "fat" because I was overly paranoid about a potentially negative connotation borking the system.

For the most part the AI seemed to understand what I meant remarkably well. (especially the "massive thighs" part) Even the rejects looked like they had potential. For example, here's a body I was actually satisfied with, that I pasted a presentable head onto, in order to replace the nightmare face it originally generated with.

I feel like we're on opposite sides of the spectrum where your AI just makes the belly and breasts big while keeping everything else thin, (Similar to Vale-City or Kipteitei's earlier stuff) where as my machine was so fixated on maintaining a certain level of realism in fat distribution it didn't make the belly quite as poofy/bulged out as I wanted, like in the reference image I provided prior.

I think we can all agree that the machine is absolutely OBSESSED with mommy milkers. It doesn't take much coaxing for the AI to go full booba-brain.

Also, sidenote: I find it absolutely adorable when the machine finally cranks out a half-decent render, and then insist on "signing" its work in gibberish.

Experiments with the furry models. Can still generate decent humans as long as you specify, and adds a lot of convenient tag recognition for prompts.

(the cat girl was purposeful - wanted to see if it could get ears and tail without going full anthro)

>>114761
>The professional artwork-looking wedding-dress waifu who just casually has 6 fingers.
Based computer-man does it again!

Anyways, I'm not super interested in generating this way myself, since I actually like chubby cheeks, but I'm super curious: how are you guys going about getting fat bodies with skinny faces?

Are you doing it intentionally or does it just kind of happen?

>>114762
In my case it's probably a side effect of inpainting the face separately with its own prompt, and forgetting to include "chubby" or similar words at that stage.

>>114769
>>114762
Here's a quick revision with a more plump face (and kind of fixed hand)

>>114771
Oh wow.
Awesome smut done quick.

>>108030 (OP)
>>114759

This may in fact, be true. Every attempt I've made to figure out how it distributes weight ends usually with failure. Now, I don't use img2img at all; I've found it to never really give me what I'm looking for. I think part of it is just that it's very... clunky? I guess? It may just be that i'm using an old webui fork rather than something like Automatic1111's.

One thing that may be causing me issues is that I mostly try to get a 'style' down; and some styles it weights heavier than others. For example, I found one artist in particular it 100% refused to depict the person in anything other than like, traditional garb. Which was a bit odd, even with having explicit prompts to the contrary. The model can definitely be finicky.

I will say that certain words, having used laionB to search up what the model is using as a basis, reveal why it really likes to make huge breasted women. Like, 'chubby' never gave me curvy women, it gave me women with breasts larger than their heads. Which is also the case when you tell it to render someone pregnant or the like. At least, with mine.

I would say that when I put out things looking for 'fat' I get something like picture 1. Which, okay, but I think people are mostly looking for picture 2 in terms of distribution? Hard to say. Attempts to actually create the right distribution are, at least for me, very hard.

But, I think that a list of good prompts may help everyone, since the prompts are the most important part for a lot of this.

>>108030 (OP)
Why the fuck is this pinned. This is the draw board. These are not drawings. They're procedurally-generated content.

When I go on this board I don't want to see derivative collages with bad anatomy created by talentless hacks who want to pretend they're artists.

>>114852
Then leave?

>>114852
Because mah free art...

>>114852

waifu diffusion superiority

>>108030 (OP)

>>108030 (OP)

>>114876

curious to know what prompt you used to get those eyes/that style of art generated. In my experience, even using something like waifu diffusion it struggles to generate that sort of thing.

>>114763

Wow that's really well done! Care to share any prompts from that? How much did you tweak images in between runs?

>>114910
Started:
((slim body)) busty cleavage,(police uniform),(pretty face),[[salma hayek]],donut shop,Michael Garmash,Justin Gerard,Greg Rutkowski
Negative prompt: naked

Ended:
((obese body)) fat,obese,bbw,chubby,overweight,(round face),double chin,big tits,sagging breasts,(obese belly),[[salma hayek]],Michael Garmash,Justin Gerard,Greg Rutkowski
Negative prompt:

Then I put the image into img2img with denoising based on how significant of a change I was looking for with CFG 15 and changing prompts to push larger and larger body

Low denoise large batch to get better hands after getting a good general image. Which I forgot to do for image 4, whoops lol

Need to delve into using inpainting to make that process better.

Photoshop occasionally to remove random junk with content-aware fill or some quick smudging

Using () and [] have been nice to increase or decrease the keyword focus in the prompt

[[person]] has worked well to get general face,skin,and hair color from a single prompt without every result looking exactly like the person

Here's some prompt keywords that have worked well
General shape: slim,thicc,curvy,chubby,fat,bbw,obese
Specific body: fat belly,double chin,pretty face,cleavage,busty,big tits

Negative prompts have also worked great as well

>>114857
This is clearly /alt/ or /ee/ material. I come here to look at drawings made by human artists. Not this low-effort unoriginal content. Regardless, it doesn't deserve to be pinned.

Niggas really be freaking out over AxelRosered tier fats because they look vaguely life-like.

>>114897

Here's a page on some anime-style prompts etc for getting better results with waifu-diffusion.
https://rentry.org/anime_and_titties
Interestingly, I've found the best eyes face was made with the "80's anime" prompt as it seems to emphasize larger and more simple facial features which can then be upscaled with more detail.
Waifu-diffusion is trained on Danbooru images and seems to respond well to prompts matching that same tagging system.

>>114952
"I clearly don't like this thing. Instead of ignoring it and moving on with my life. I will go and complain." Please stfu.

>>114952
Dawg, there's 1 stable diffusion thread and like 30 human art threads, nobody's stopping you from posting there instead.

Imagine calling this thread "unoriginal" it's the 1st of its kind.

>>115102
The art is unoriginal. It is derivations of patterns from existing artwork. That's why you guys tell the system to copy the artstyle of existing human artists instead of making something unique.

I don't think non-human art should be on this board, but if I saw the thread normally I would have done what you said and moved on. However, it is pinned for some reason. It clearly does not merit being pinned. Now I am forced to see it every time I open the board.

Can you give me a skinny, lanky girl with black bangs covering her eyes, and swoops like meryll. In something skimpy like a crop top and shorts. I’ve wanted an easy ref for an oc for a while… (stuffing-based so if you want to give her a massive gut, sure.)

>>115187
bruh

>>115187
Even if you couldn't hide the post like >>115241 suggested you can just learn CSS and/or JS and hide it yourself
Basically any issue you have with a website can be fixed with enough determination/stubbornness

>>115198
Here anon, take it

>>115454
Ey, thanks.

>alright let's crop down to just her head so that the AI can focus on rendering a face
>The AI:

>>115812
mind sharing what kind of prompts you used for these?

>>115822
They were made with "yiffy" YE18 finetuned model. While it's biased towards furry art, it also has a lot of training on general NSFW content and tags. I borrowed the prompt from someone else and modified slightly to generate non-furry results. Still get the occasional animal ears or purple skin though.

Prompt:
beautiful human woman. uploaded on e621, questionable content, by einshelm, an extremely talented impressionist painting of a plump maiden, curvy, soft, squishy, pudgy, belly, midriff, ((impasto impressionism))

Negative prompts:
poorly drawn, bad anatomy, missing limbs, extra limbs, deformed, disfigured, warped, misshapen, mutated, mutation, penis, (((nighttime)))

>>115834

Does this work on any phone app vertions, or what site or app that is being use to make them?

>>115834
Those negative prompts are actually funny to me. Man's literally telling the robot to draw better.

Does that actually make a noticeable difference?

>>113222
Inpainting

>>115844
I didn't touch the negative prompts from the version I borrowed, but in my testing it does actually help a little in some situations. The AI can make a child's crayon drawing as easily as a professional looking piece, so it can sometimes help to narrow down what you're asking for.

>>115843
I ran it locally, and haven't kept up to date with website versions. Since I'm using a community finetuned model, I wouldn't expect the same results unless the site can be made it use it somehow.

I have only seen this stable siffusion thing for the first time last month on this thread right after it was created.

And in this month shit has progressed so fucking fast i fucking swear it almost scares me.

An alluring plump young woman, fat belly, wide hips, thick thighs, wearing a tanktop and skirt, full body portrait, beautiful, intricate, elegant, bright daytime lighting, 8k, classical painting, photorealistic lighting, matte painting

Scale was 10 and steps were 30. I used liquefy in Photoshop to make an initial output fatter. sent it through again to make it coherent. I cropped and reprocessed any part I needed to, just the face here.
I use REALESRGAN to upscale my images, I also use a tiled upscaler when I feel fancy.

Welp, I've been replaced.

>>115858
Third one looks like a squash lol

>>115859
I know lol. All I did was turn up the denoising strength by about 0.5 and the machine was like: "Did someone order pudding?"

I'm thinking maybe I'll Frankenstein one of the portrait shots over some of the faces rendered in full body generation. Through photoshop since I've had terrible luck with inpainting.

The zombie eyes on the last one are kinda cool; but definitely not what I had in mind.

hmmm....?

Im finally getting somewhere with this

NovelAI

It's really something else

>>116106
This stuff looks great!
Is there anyway to do a free trial on Image generation with Novel AI?

I've tried so far and keeps telling me to get credits, so I'm wondering if I need to take the plunge...

>>116149
I'm pretty sure you have to take the plunge, unfortunately. The power of this thing is amazing though.

Protip for anyone else using NovelAI: Using both 'fat' and 'obese' as tags tended to muddy the results. It's better to just pick one rather than spamming tags.

Been playing around with stable diffusion the past couple days. I've started with different prompts and models and been grabbing anything that comes out okay and moving it to a keepers folder. Some results

>>116149
Nothing that's free in terms of money is flexible enough to be worth the time. But I use colab and even after the payment changes they've made it is still a good deal.
About 100 hours of SD runtime for 10$

Old raw output

>>116382
These are quite good!
What does your prompt look like to get these?

>>116622
On mobile rn so I don’t have access to my prompts atm. I really liked the style that
>>115834
Was able to achieve with the yiffy ye18 model. I went and installed the model and started playing around with different prompts. Started with what previous anon did and modified a bit. Key prompts I added that helped were
Positive: reclined in bed, horny, fucking, gushing, sweat
Negative:((tail))

>>109781
>what if people where seals?
Mermaids as aquatic mammals lol

Here's a small sample of what I've created so far

одета в школьную в форсу, с мечом

https://www.deviantart.com/arg343/gallery

Experiments combining Stable Diffusion art with captions to make sequences/erotica, lmk if you think this is cool or stupid

anyone here know why this google colab stops working after more than one or twi generations?

https://colab.research.google.com/github/pinilpypinilpy/sd-webui-colab-simplified/blob/main/Stable_Diffusion_WebUi_Simplified.ipynb#scrollTo=gk1TyBA0Arxt

and if not is there someone who knows another one i can use?

>>116903
Here's mine.
Needs models in your Google drive.
Has most of my goodies, mask maker from centipede diffusion, crop tool, and a tiled upscale tool from stable diffusion batch.
I might swap it over from deforum to fast SD but I haven't seen it in action yet.

https://colab.research.google.com/drive/1TM7g37ZpxTm8Qdv3KsJj42-3s2a2TMVS

Big boobs, nipples, cute face, goddess

Huge boobs, thick legs, nipples, cute face

Huge boobs, thick legs, nipples, cute face

>>117041
is there any video teaching how this works?
or can you perhaps give me a description of what to do?

yeah i'm that stupid

>>117087
The whole first section is setup. The checkpoint files are meant to be in the path AI/models/ in your Google drive and needs to match the filename in the "select and load model" section.
The animation settings cell will need run at least once.
There's a prompt in the prompt section. Along with a small explainer on how it works. You can just modify the text in the quotes.
Init settings are your img2img and variations. Set use init to false and the strength to 0 if you aren't using it. Keep the strength at 0 for variations, and raise it for img2img.
Outputs aren't sent to Google drive by default but can be if you want, that's in the model and output paths section.
The resolution should be kept at 512x512 or a resolution with the same number of pixels, higher resolution will make unexpected repetitions, small ones get deep fried at higher step counts.
REALESRGAN can be used to upscale your images and those will be saved in the same directory as the batch folder.
using the tiled upscale on the results from REALESRGAN helps keep the style in tact, but make sure the image width is divisible by 512 for that section.
The cropper and mask tools are pretty self explanatory, but the cropper requires the image to be on your local machine.

>>117098
I heard the checkpoints and model were leaked for NovelAI, reddit is doing its thing and making actually looking into it impossible. Any pointers?

>>117155
i really don't wanna sound this dumb but i'm a totall newbie here.

your explanation doesn't really help much sorry for the trouble

>>117286
i have no idea what a checkpoint is in this context, nor how to download the model. i mean in the beggining of the text it talks about stable diffusion and then K.diffusion. i'm so throughly confused.

>>117155
> Check 4chan /t/ board

>>117286
>>117289
the checkpoint files are the weights used by the diffuser. You get them from the huggingface website, it'll require an account. Google is your friend here. Then get the files on your google drive in the location previously mentioned.
Run the cells one by one and read what they do.
K diffusion is just the implementation of the diffuser architecture using Keras, a high level AI programming language.

>>117332
>4chan /t/ board
You are a friend of the cause, thank.

>>117433

No probs! Hope you're enjoying it!

On topic: I've recently had a friend let me use NovelAI's website, and I find its image generator to be quite limiting in comparison to Stable Diffusion. However, there's no denying the ease of access and capabilities of it. SD images feel more lively and less... generic compared to NAI.

A pretty nice painting generated of a pregnant fat alluring young massively pregnant woman, I think this is one of my best so far of that style.

This pain is just too real.

Девушка с большой грудью

Maid asses

Ruthless. Business. Woman.

...and then the machine added a little heart. What a charmer!

How do you use this?

>>118701
Where do you download or use it?

>>118702
Read the fucking thread

>>118777
I’m genuinely sorry if you’re having a bad day.

I did find the links to what modifications other were using shortly afterwards so I’ll get to posting any results, I already was reading it, the question was in case I couldn’t, I’m sorry for any inconveniences.

>>118804
I meant “others” plural for the record. I can’t delete to change it, sorry again.

https://onceuponabot.com/story/486a3498-c6a8-42a4-b071-e1ebb9bff268

Thought you guys might want to play a bit with this

>>119253
https://onceuponabot.com/story/725d0f82-8ea6-4053-9cf0-f2bae40dd885

damn this site is fun (make sure to clear cookies when you run out of free generations.)

>>115454 counter requesting to make the rest of the fat

>>115454
hey chief, what prompt did you use?

Little girl

>>122590
The hell is wrong with you?

BBW