Previous Thread: >>137349
>NAI Leak Speedrun
https://rentry.org/nai-speedrun | https://rentry.org/sdg_FAQ | https://rentry.org/sdhypertextbook

>/vtai/ Resources
Holomem Prompt List: https://rentry.org/3y56t | VTuber Model Mixes: https://rentry.org/8nxtk | Embed/Thread Archive: https://mega.nz/folder/23oAxTLD#vNH9tPQkiP1KCp72d2qINQ

>/hdg/ FAQ + Embeds, Hypernetworks, Models
https://rentry.org/hdgfaq | https://rentry.org/hdglorarepo | https://rentry.org/hdgrecipes

>/hdg/ LoRAs
https://rentry.org/hdglorarepo | https://rentry.org/59xed3 | https://gitgud.io/gayshit/makesomefuckingporn#lora-list

>Local Install
Nvidia: https://rentry.org/voldy | https://github.com/AbdBarho/stable-diffusion-webui-docker
CPU: https://rentry.org/cputard | https://rentry.org/webui-cpu
AMD: Native: https://rentry.org/sd-nativeisekaitoo | Docker: https://rentry.org/sdamd | Onnx: https://rentry.org/ayymd-stable-diffustion-v1_4-guide

>Cloud Hosted Install
Colab: https://colab.research.google.com/github/TheLastBen/fast-stable-diffusion/blob/main/fast_stable_diffusion_AUTOMATIC1111.ipynb | Paperspace: https://rentry.org/865dy
Colab for Complete Retards: https://colab.research.google.com/drive/1STL60qfoY-iSlhRb9zFETRLTqhNbznRf

>Subscription sites
https://novelai.net/ | https://nijijourney.com/en/ | https://yodayo.com/ | https://holara.ai/

>Guides
Prompting: https://rentry.org/hdgpromptassist
Inpainting/Outpainting: https://rentry.org/drfar | https://rentry.org/inpainting-guide-SD
Upscaling: https://rentry.org/sdupscale | https://rentry.org/hiresfixjan23
Animation: https://rentry.org/AnimAnon | https://rentry.org/AnimAnon-Deforum
Wildcards: https://rentry.org/NAIwildcards
Embeddings: https://rentry.org/simplified-embed-training
Hypernetworks: https://rentry.org/HNSpeedrun
LoRAs: https://rentry.org/lora_train | https://rentry.org/HDGLoRaIssues

>Old News
https://rentry.org/sdupdates3

>GPU Performance Data
https://docs.getgrist.com/3mjouqRSdkBY/sdperformance

>Upload your images
https://aibooru.online/
https://www.pixiv.net/en/

>More Resources
4chanX Catbox userscript: https://rentry.org/hdgcb
Essentially everything: https://rentry.org/sdgoldmine
Tags: https://danbooru.donmai.us/wiki_pages/tag_groups
/g/ wiki: https://wiki.installgentoo.com/wiki/Stable_Diffusion
5ch NovelAI wiki: https://seesaawiki.jp/nai_ch/
Models: https://civitai.com/ | https://huggingface.co/ | https://rentry.org/sdmodels
More links: https://rentry.org/sdg-link

Credit to /vtai/ for the template
And, please, put photorealistic gens in /ee/ or something. If we wanted realistic girls, we would be on a different board

Between getting decent at this and having access to Eleven Labs maybe I'll try making some voiced content to go with some AI generated characters and have them read some scripts.

>>9 (OP)
If you need help, check out the above, or the following posts, to see if your question has already been answered:
>>141927
>>141956
>>142362
>>142008

LORA
https://civitai.com/models/6638/samdoesarts-sam-yang-style-lora

I'm confused, what are some of the tags used to make some of these pictures because when I do it they're not as big as the ones on this thread

>>14
what model are you using? Normal models have very little fat training and can't do it well. Try using biggergirls or the model from the last thread and get back to us
https://civitai.com/models/2664/bigger-girls-model

>>13
Cute, especially the 2nd pic

>>14
I used a model that merged Bigger girls and Koto. Here were my tags:

masterpiece, best quality, highres, undersized clothes, big belly, long Hair, blue eyes, curvy, wide hips, obese, hand on belly, hand on stomach, tube top, miniskirt, thighhighs, black clothing, choker, belt, belts, garter belt, black belt, nipples, classroom, teacher, glasses, chalkboard, desk, converse, shoes, 1girl, fat, bbw, huge belly, apple, red apple, in frame, full body, seductive

and here are my negative tags:
ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, bad anatomy, watermark, signature, cut off, low contrast, underexposed, overexposed, bad art, beginner, amateur, distorted face

>>15
NovelAI

>>16
I copied your prompts all of them and the girls in the pictures i get are not nearly as big as yours, im confused, btw. what is Koto and how did you merge the two?

>>143472
Could you upload the 4th and 5th picture here https://catbox.moe/ and post the links for me please?
Cross posting my request just in case.

>>19
https://files.catbox.moe/92nkks.png
https://files.catbox.moe/hj6dfo.png

>>19

oops I miss counted
https://files.catbox.moe/fpneww.png

>>20
The question being how does she even got to wear a cap like this?

>>23
I bet it's actually a beanie with hard visor

ControlNet is pretty bad on the canny and depth modes. OpenPose mode is insane if you can get it to detect bodies.

Another example. Needed specific prompts to keep the pose.

>>18
>>18

Koto is this model from Civit AI
https://civitai.com/models/5245/kotosmix

You can merge two models in this menu pictured. I forget if this menu came from an extension or not. Just in case, here is a list of my extensions too.

>>27
Thank you! Also, are you the one that used that Lora? How do you use those?

Posting it again in this thread
ControlNet
>https://www.reddit.com/r/StableDiffusion/comments/119o71b/a1111_controlnet_extension_explained_like_youre_5/
OpenPose
>https://www.reddit.com/r/StableDiffusion/comments/11cuztu/openpose_controlnet_preprocessor_options/
OpenPose Editor (get it through the extensions tab)
>https://github.com/fkunn1326/openpose-editor/blob/master/README.en.md
Random video that uses OpenPose
>https://www.youtube.com/watch?v=iFRdrRyAQdQ

> pov: you caught Mio on her third post-workout snack
+ assorted pics

Still trying to get a handle on poses.

>>9 (OP)
Hey what tags did you use for that Nina kosaka pic

>>34
"(masterpiece, best quality, highres, detailed:1.1), <lora:allVtubersLora_hll31:1>, nina kosaka, fox girl, fox tail, multiple tails, facial markings, red eyes, black cocktail dress, cleavage, straps, (fat:1.1)"
and negatives were
"(worst quality, low quality:1.4), blurry, depth of field, (motion lines, blur:1.1)"
I used the vtuber lora from here https://civitai.com/models/10664/all-vtubers-lora and inpainted the face with anythingv4

>>32
What LORA are you using for the first picture?

>>27
Different person here, really struggling to get anything considerably big. I don't know a lot about loras, but I merged bigger girls with hll-epoch-00015 for vtubers and was struggling to get anything big. Tried copying settings as best I could but it didn't quite seem to work.

>>37
you're going to want bigger girls in the A slot on the merge. right now it's 70% All VTubers and 30% Bigger Girls. You're probably getting too much of the Vtuber model.

>>30
Here is a tutorial I used to start using LORA. Basically, you use a LORA trailed model and inject it into the base model that you are using. I'm currently using this LORA:

https://civitai.com/models/6638/samdoesarts-sam-yang-style-lora

But results are mixed. Going to try some different ones soon.

NOTE: you need to be on the latest version Stable Diffusion to use LORA natively. otherwise you have to install it a different way that I don't know anything about.

>>37
Using the lora wrong. You're supposed to put it in your models/Lora folder and select it using the "show extra networks" button. It's better this way so that you can get the effects of the lora without watering down your model.
And merging with normal models will usually make girls, smaller, yeah, since most normal models have little to no training on fats. You can use more emphasis like (this:1.2) on your fat tags, use a merge with more bias towards your fat model, or not merge at all and use loras for characters instead. Still looking good though
>>36
I'm only using the vtuber lora. My model is here >>142008 if that's what you meant

>>31
Controlnet is so goddamn cool. I heard of 'latent coupling' which I dont really understand but supposedly lets you put multiple prompts in one image. What else have you been able to make with it?

>>29
What kind of settings and prompt placement are you using for your LORA?

>>40
Any tips for getting it to look...better? Kotos on its own works pretty well, especially with one of the Calliope Mori Loras I used now that I know where to properly put them, but, I'm still getting a large quantity of really bad ones. Ideally I'd really love to get some that are like the bigger ones you do, or even bigger.

Also just figured I'd make sure, inpainting with anythingv3 is the way to go for fixing faces when you have a shape you like?

>>43
I got a few tips:
1) don't go above 512x in either dimension before you upscale. Models are trained on 512x512 images so when you go any larger than that the ai doesn't really know what to do and is way more likely to fuck the image. If you want a portrait do something like 384x512 or 448x512 and then use a larger upscale like 1.75 or something
For example, img#1 of this post is a 1024x without upscale and img#2 is 512x upscaled to 1024x.

2) Don't use euler a. It's pretty outdated (that 'a' stands for 'ancestral'!). Use basically anything else. I like DPM++ 2S a Karras on 25-35 steps, but I also haven't used most of the samplers, so take that with a grain of salt

3) Use a VAE. If your colors are desaturated 90% of the time it's because you have no vae or it's not being used. The big ones are either the anythingv4 vae or the waifu diffusion vae. I'm on anything v4.
https://huggingface.co/andite/anything-v4.0/blob/main/anything-v4.0.vae.pt
https://huggingface.co/hakurei/waifu-diffusion-v1-4/blob/main/vae/kl-f8-anime2.ckpt
You put it in models/VAE and then select it from the settings menu. Do not use 'auto', it's broken, explicitly select the vae you want
Img#3 is a comparison of the vaes

And yeah, inpainting is cool and good. If the fat model is having a hard time, sometimes i'll switch to anythingv4. In img#4 i switched to anythingv4 and inpainted her face with 'masked only', max padding, and a resolution of 768x.

>>33
What on earth was the prompt/setting for those first two?

>>44
Didn't even know my VAE wasn't enabled, I had installed one and threw it in that folder but didn't know what was going on. Ended up baking in anythingv4 VAE into my bigger girls and Kotos mix because the colors were washed out thinking that was the solution, thanks!

And, any advice with inpainting? More often than not I get image 1 over image 2, and even when I do get image 2, they turn out...not the best (think this was best of maybe 12 that had just her face? Ran a few batches)

One last thing, any advice for more bigness other than just messing with models? I'm in love with how yours usually seem to turn out but I'm getting lots of different shapes and not much is that appealing, really. Is it lots of trial and error, and a few loras? Image 3 is how most batches turn out, it's rare that I get anything really good

>>45
https://files.catbox.moe/zzk4yh.png
https://files.catbox.moe/gaci5d.png
I based an openpose off of a roundersofter image. The bones are placed pretty poorly.

I usually avoid the realistic models, but biggergirls irl seems pretty bad.

>>21
>>22
Thank you.

>>44
Could you catbox that 4th Nina please? I don't think you posted it on vtai.

Are we allowed to make requests?

>>46
That inpaint issue you got is because your denoising is too high or your padding is too low. Try a denoising of no more than .6, and max padding. There's a more in-depth guide here >>142362 but 0.5-.6 & max pad almost always work well

And if you have trouble getting something specific, it might help you to go through the tags for your model. Biggergirls comes with an artists & tags list that i've attached, go through it and it might give you an idea of what to use. But base BG doesn't do wide & thick very well so I'm using the version here >>142008 instead

>>49
yeah
Your other stuff is cute, at least

>>50
https://files.catbox.moe/eplld9.png
Damn I wish the userscript worked here

>>51
Yeah sure, just ask nicely

>>32
can you share the tags for the first 2?

Tried to copy the bws panel.

Giantess. Probably from combining "from above" and "massive" with the river setting.

>>52
Ok ty.
Is it ok if someone makes one of the mario princesses with ahegao and nude. Or just vare breasts in general?

>>52
how many vtubers does your Lora or whatever have? which companies and does it have indies? it's just it would be foolish to request someone you don't have at all haha

>>143938 yeah wrong thread, it's a VAE not a Lora model in this thread, anywho, question still stands

>>143940
i was asking
>>52
how many vtubers his model could do, he showed nijisanji and hololive, but if i requested an indie it might be impssible. anon also said he would do request if asked nicely.

Sorry about the image dump. Ran most of these overnight to see what I got.

>>52
Thank you.

>>58
Here's some stuff of Rosalina. She's my go to Mario Princess.

People seem to make Loras for everything.

>>59
I actually have no idea. It supposedly understands vshojo, hololive, & niji, but certain characters are completely unrecognizable (like, they even included dudes, but not ironmouse? are you serious?), and the page doesn't have a list of tags. It apparently has some indies although the only one i've tried is shylily and she came out alright

>>53
https://pastebin.com/ewyajkeP

>>70
forgot the shylily & lora link https://civitai.com/models/10664/all-vtubers-lora

It can almost do a sequence but it usually just makes them the same size.
Also: Baemia

Tried out the default openpose position with minimal prompts. Not perfect, but workable.

>>74

>>71 if its not a hassle, could you try if buffpup exists? if not i would also love a shylily that's bigger than this one, looks decent. thanks for the lora

>>76
I downloaded the shylily lora and gave it a few gens. I don't know who this weird inner-tube creature is and I do not care to learn.

>>67
Thank you alot also rosa is the best girl. I just wanted to leave it open so people could make what they wanted. Is it possible to do proper ahegao with tongue out. Just curious. I dont know much about this but find it interesting

>>76
Tried buffpup but didn't work. I got shylily instead

>>77
Cute gens. I also had no idea there was a shylily lora, it works pretty well. It inpaints like shit though, have to turn it off to do fixes

>>81
I'll set the render to go overnight and see what I can get

Berry, alien, or slime?

Some of these are as large as bgv1's immobiles.

>>81
The stuff overnight didn't work so well, but I ran it again this morning with fewer batches and got a couple good ones. Here are both

>>81
>>81

>>81

I forgot how cute rosa was. Nice posts all around.
>>68
> People seem to make Loras for everything.
God damn yes they do. There's a rosalina lora that also does alright
https://civitai.com/models/16316/rosalina-or-character-lora-1077

I have more I will be posting later. Anyone know how to make what happened in the fifth image not happen? The multipanel thing managed to infect a ton of images in my later generation runs.

>>97
I managed to exceed the size limit with the fifth image. Here it is by itself.

>>98
You could try a combo of controlnet pose references, emphasizing a 1girl prompt, and negative prompts for comic and panels.

could also be a seed issue if you're using the same seed.

>>97
"solo" helps, but it usually makes the background blank

>>99
With controlnet openpose, you can skip those prompts. Additional characters sometimes appear with weird poses or prompts.

Kronii got fat.

Be great if there was a ai image of a fat naked Mary from walle

Doing some preliminary testing with controlnet open pose. This absolutely rapes my gpu/vram though.

>>104
Yeah mine too. I get slightly better results by ticking the 'low vram' box but it's still slow as hell. I'm probably gonna avoid using controlnet

>>105
It's much faster if you don't click low vram, but you are significantly restricted if you don't have over 8gb of vram.
All of these were generated using the bbw model, or a cleaned version of the bbw model saved in fp16

>>109

Wow. This has got to be some of the best art I have ever seen. I'm impressive

https://civitai.com/models/8949/jim-lee-dc-comics-marvel-style-lora
I get nicer results with lower denoising than usual on the latent upscale on this Lora.

Fellas, is there a good ai for dubbing bbw comics and whatnot?

>>112
The only thing that really comes to mind is eleven labs but I think that's a paid subscription

>>113
It's $5 a month for like 30,000 words. Plus you can generate your own voices. First month is only $1.

The free version has a bunch of good stock voices, but you only get like 3,000 words a month.

Goofy blob pose. Stack the hips, knees, and ankles together.

without the comic book lora

>>113
Thanks anon, worked like a charm. Now all I need are some stock burp sounds.

>>117
Can anyone tell me which AI program is used? It's a page? or is it directly a program that is downloaded?

>>121

>>141927
>>9 (OP)

>>120
Check out this program:

https://getsoundly.com/

I use it all the time in my sound editing work. The free version comes with tons of SFX that are easily searchable.

I think this is my best picture yet.

Random stuff.
>Img2Img to get a b&w image
>Canny controlnet to color it in
>Img2Img to make it nicer
>Img2Img to upscale from 512x512 to 896x896

Regular open pose stuff.

https://bbw-chan.nl/bbwdraw/res/137349.html#141927

>https://huggingface.co/stabilityai/sd-vae-ft-ema-original/resolve/main/vae-ft-ema-560000-ema-pruned.ckpt

Most vae files I find are .pt and this ema vae is a .ckpt, which makes me wonder if it's really a vae or just another checkpoint? anyone know?

>>128
It's a normal vae, you just have to put it in your VAE folder for it to be recognized

This thing is pretty good...

>>129

That's what I did, I just had trouble telling if it was overwriting the builtin vae on my main checkpoint or not and I was wondering if I needed to alter the file extension too or whatnot

>>130

wow amazing wet effects! AND somehow 2girls despite the ai almost always getting confused when 2girls touch each other, very impressed

99.9% of the sounds in the free version tell you to upgrade to pro in order to export :(

was really hype until I noticed that

>>132
Yeah, alls I mixed was biggerGirls and anything v4 (3:7) and then that with Kotos (3:7). No baked vae, and it sort of takes whatever one you choose even though biggerGirls generally freaks if you donʻt give it the anything vae. Then I gave it the 2girls and group pose prompt.

>>49
How do you guys make this?
Is this all from the same page (NovelAI), or are there multiple sites for making AI art?
Sorry if I am bothering, I dont know much about this subject.

First time playing around with the NovelAI leak

>>135

I've had the most success with google colab. Here's a copy I made of the one from AUTOMATIC1111. Follow the FAQ from https://rentry.org/sdg_FAQ#how-do-i-setup-the-leaked-novelai-model. You will need to add the animevae and hypernetworks models.

https://colab.research.google.com/drive/1duniuatVisudX25R0Qz7vWwMr59-fSI6?usp=sharing

good results with low pov, looking at viewer

The BiggerGirls model likes to turn every thing into cameras

Also, wet gets some interesting results.

Adding "belly button" to the negative prompts ends up with a more subtle look.

Here's some after a checkpoint merger of novel ai full and bigger girls v2. The results are a bit noisy, but you can have more things in the scene.

>>143
very cute!
>>145
The grain will go away pretty easily with an upscale (if you're able), 512x512 is just too small an image

>>138
Could you catbox the 5th image please? 095.jpg?

>>146
Pardon,Just curious as someone thats not knowledgable about ai art how much work went into the first image?

Trying and failing to upscale, but sharing anyway

I tried and failed to download and set up the NAI leak but I'm too boomer to understand GIT. So can I make a request for naked princess peach ahegao like the rosa ones?

>>146
The next time you do a regen/merge of bbw-step-30k could you use HoloKuki as a base? It has much better background, and allows for much higher cfg scales than Anything v4. It's also much more compatible with various embeddings and LoRAs. Definitely something worth looking into if you are looking for a new model to base the step model off of.

>>137

can you clarify for me please, is novelai just another checkpoint that competes with things like biggergirls anythingv4 etc etc? or is it something other than a leaked professional checkpoint?

>>152

Its the cktp used for NovelAI.com. Its probably out of date to what they use now, but its the best (imo) general purpose anime generator. Like NovelAI knew what to do with "carnival" and "crowd of people". whereas biggergirls seems to have lost some of that with their focus.

Supposedly Anything v3 is based off this leak, but I haven't been too impressed with my comparisons.

Still tweaking my merge, it would be nice if I could make a "run and gun" Google Colab file that would just "work". For example, the one above breaks doing some of the extras (I've fixed on another private file).

>>152

Some words:

The first two images I posted in this thread are from the NovelAI leak unmodified. It's an uphill battle to generate anything but small waifs or beach balls without curves.

Next I tried the biggergirlsV2 cklt. The issue there is that if you wanted a scene or props, it often failed. I personally feel the anatomy is a bit off, but that could just be my prompting.

Stable-diffusion-webui has a feature where you can merge ckpt files. So I merged bgv2 with NovelAI @ 0.65. I may try bumping it down to 0.5 to increase bgv2's weight. But even at 0.65 if I don't mention weight at all I still get some thick middles on like 3 / 10 images.

I tried merging bgv2 with Anything v3 and it just wasnt as good. I know bgv2 is based on Anything v3 so there might be some stacking weights going on - especially with the faces.

>>148
This has more or less always been my workflow >>142362
Generate many images, pick my favorite, upscale it, and fix any errors with liberal inpainting. Doesn't take all that long but depends on luck and how difficult the prompt is. That 1st image in particular actually took closer to 30 cause I kept rolling shit faces & hands on the inpaint, but if you aren't as picky as me you could easily do it in 5-10

>>151
I will absolutely look into it. The hll mixes are pretty and have the additional benefit of not needing a lora to generate most vtubers. Although, my concern is that no matter what model I choose to train on fats, I'll still get low quality output because most fat art (and by extension my dataset) is low quality.
This is especially apparent when you try to compare a non-fetish pic from a fat model vs a normal model- the fat model gets blown the fuck out pretty much every time. I want to tweak my training settings, or try making loras, but it'll take some time

Here are four comparisons.

1. NovelAI Leak 100%
2. Novel AI 65% - biggergirlsv3 35%
3. Novel 30% - BG3 70%
4. biggergirlsv3 - 100%

Just as a comparison - the only FA keyword thrown in was "bloated". IMO, bgv3 makes it harder to fine tune the fatness overall. It caters to more extremes. On the flip side, the NovelAI leak makes it difficult to make anyone more than 50 Lbs soaking wet.

The two in the middle are the merges I mentioned from stable-diffusion-webui's Merge tab. Personally (deformities aside) number 2 sits better with me. I can always throw in more emphasis braces or add more fat descriptors to increase weight.

Nothing against the other models if that's your cup of tea. Just trying to crave more of a niche in the AI space to get things to stand out a bit more.

Here's one I massaged in the #2 ckpt

>>147
https://files.catbox.moe/nwaqsp.png
That one is just a lucky seed, controlnet was off at the time.

>>158
Thank you.

Have also been playing with prompting for just a regular scene and then tweeking in img2img. This is just using the NovelAI ckpt.

Got the thin one on the left, then moved into img2img, running about 20 queries with 10 batches. Luckily the open shirt was a pure accident.

The best approach is to start off without any weights in the prompt and then increase things from there to get a composition you can live with. Then take it to img2img and work on molding.

Want to experiment with some inpainting next.

https://civitai.com/models/15773/pixar-style-model

The pixar themed model is cute but it only does faces and it would take a large area for the details to come through on inpainting.

I've got a question more for OP since I have been wanting to experiment with a lot of things they've been suggesting after seeing it in the last thread and this one, but how do you go about making sure the Latent nearest exact doesn't always end up with pixelated results? I've never really used the setting before and even with the same exact prompts and models I still find it getting heavily distorted around certain parts. I can understand it being a hit or miss sometimes, but it feels like it's hit or miss a bit too much for me at least

And on a bit of a side note, do people still use the Eta seed? I've found just setting it to 0 gives a lot more variety, but I guess can sometimes give a bad variable that doesn't work or makes certain seeds inconsistent. Wish it would display in the output files honestly

>>164
Continuing off of this post, since I haven't had too much luck with Latent (Nearest Exact) I've been using ESRGAN_4x for upscaling the images I've found to be the better outputs. I normally keep it on between 0.6 or 0.7 depending on how much detail I want saved.

I also completely forgot about sending the prompt and other stuff that probably matters too for anyone that wants to try replicating this.

Prompt: (masterpiece, best quality:1.2), <lora:hyperfusion100k_v4:1.2>, cat girl, cat tail, cat ears, inflation, immobile, bloated, latex bodysuit, (gigantic body inflation, inflated arms, inflated legs:1.1), (puffy cheeks:1.2)

Negatives: (worst quality, text, low quality, multiple characters, multiple views, clone, clones:1.4), error, blurry, depth of field, speech bubbles, out of frame, cropped, monochrome, jpeg artifacts, signature, watermark, username, artist name, empty background, ocean, 3d, simple background, cum, cum on body

Seed: 1189432184, Sampler: DMS++ 2S a Karras, Steps: 25, 512x512, 1.75x upscale resolution, AnythingV4 merged with 80% Bigger Girls V2, plus 1.2 of this LORA (seen in the prompt) https://civitai.com/models/16928/hyperfusion-100k-images

>>164
>>165
> how do you go about making sure the Latent nearest exact doesn't always end up with pixelated results?
Depends heavily on your prompt, model, loras, and resolution. I don't know the intricacies, but for latent upscale, what I've personally found is that:

Tags with less training artifact harder. For example, in my model, "fat, hose in mouth" artifacts earlier than just "fat" because 'hose in mouth' does not have many training images. This also applies to loras and how they've/their tags have been trained, and can also be more apparent when you use emphasis on a problematic tag/lora

Different models artifact at different points than others, presumably because of their training, dataset, and merges

Higher resolutions artifact hard. Upscaling above 2x is unusable, and a pre-upscale resolution above 512x is finnicky. If you want a higher res than that you should consider using the upscale script in img2img (although this isn't your problem, looking at your settings)

So, shitty latent upscale is just the price you pay for using that model and lora. Besides making your own model & lora, your only real options are to use a different upscaler or up your denoising, which you've already figured out

> do people still use the Eta seed?
not really? It doesn't do anything that controlnet/img2img/lora/higher cfg scale/emphasis/better prompting doesn't
> Wish it would display in the output files honestly
Every image you generate is embedded with its generation data, like its seed, model, prompt, etc. Not normally viewable unless you have the img browser extension, which is really nice to have. But you could also just google an img metadata checker and upload your image and see its info from there

>>163

what do you mean it only does faces? the clothing looks pixar style to me? or are you saying you asked another model like anythingv4 or whatever for a pixar style outfit then only used the pixar style model for the face inpaint?

Been getting into loras.

>>168
What lora did you use here? Or was this one you trained yourself?

>>168
Could you catbox 2, 5 and 6 please?

Started up a DA for the shit I've been making:

https://www.deviantart.com/fulforget85/gallery

Anyone have any good cat girls?

Only NovelAI. Found that if you base your prompt around "small belly, early pregnancy" you can tweak in img2img to flesh out the output a bit.

>>172

Just for you anon

>>172

and another cat

Thick Thighs Save Lives

>>178
amen

Leg is a little messed up, but, I thought this Calli turned out pretty well. Also, a Shenhe.

>>179
Also, I've been adoring every Nina you put out. They're all wonderful

>>180
Could you catbox that Mori please?

>>183
Now you have to do stocking, it’s principle of the thing. I don’t make the rules

>>177
Cute, but one thing though make her bigger

This is one way to get $$$ from the ATM

>>187

Emberassing.

>>185

>>180
That IS a nice calli, and you got me inspired to make a few. Prompt? And I'm glad you like the Ninas :)
>>183
came out super well, esp #5, how'd you get that?

>>190

Those came out great! Here's what I used for my prompt, it was pretty basic. I think the Lora did most of the work because it's Calli specific.

Prompt: best quality, hires, high quality, pink hair, red eyes, calliope mori, <lora:calliopeMoriHololive_v10:1>, obese, gigantic ass, huge thighs, wide hips,

Negative prompt: (lazy eye), (heterochromia), lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, bad feet, extra limbs, (multiple navels), (two navels), (creases), (folds), (double belly), thin, slim, athletic, muscular, fit, fat face, blemished stomach, rash, skin irritation, blurry, bad face,

Steps: 30, Sampler: DPM++ 2S a, CFG scale: 8.5, Seed: 1152647543

>>182
https://files.catbox.moe/1ltcs8.png
What's the point of catbox'ing?

>>192
Thank you. As for why, iploading your generations to catbox allows other users to see your prompts and settings. It's the most convenient way to share prompts since you no longer have to copy past everything yourself. Catbox'd images also allow WebUI to read the metadata directly on the PNG Info tab so it's convenient for the receiving person as well.

I think I'm getting the hang of this

settled on this beauty

>>190
Can you tell us which model you used?

>>9 (OP)
>>184
You're right, here's some Stocking

>>179
that elira is great, inspired me to make some bottom heavy eliras.

>>198
Appreciate it. That's a fine elira you got yourself, got more?

>>196
Custom >>142008

Could somebody try to make Graf Zeppelin from Azur Lane? I'm still trying to figure out how to use the AI. Make her as big as you wish :)

how can a beginner start with ai? id love to do something with cindy from full night nanny but I had no idea where to start or what to do. how does ai even work?

>>201
Best bet, follow the "novel ai speedrun" guide in the OP. This will walk you through downloading the software and a basic model and thats basically it. Then follow the guides also linked In the OP for prompting, inpainting etc and they'll show you the ropes. Once you've done this you're good to go as a begginer. Have a play around then you might want to get better models such as those linked the /vtai/ resources links in the OP. Just remember those models (since they come from the vtuber board) are specialised for making vtubers. But don't worry about that for now, just follow the novel ai guide and then use the guides in the OP to learn to prompt and feel free to ask for any more advice.

>>198
Could you catbox that please?

>>199
I'm not the one who asked but that gofile link is dead so he can't get it, just so you know.

>>203
sure!
https://files.catbox.moe/sk079p.png

>>199
yeah, here's one of the wider ones from that batch.

>>204
>>205
Thank you. And great stuff, by the way.

>>172
Been needing an excuse to generate based dumb pink cat girl, and an excuse to try generating a sequence again

PoV, your cooking ruins her career as a swimmer but she's afraid to admit the parts she's enjoyed of gaining weight and over exaggerates the negative emotions instead when in public

Also prompted some chubby elira in an onsen. Not sure why but I love having these detailed backgrounds.

>>203
ahhh god damn it
whatever, the mega is better anyway
https://mega.nz/folder/WFk13apa#WvCEO8IoqgSaeoKQILeIqg

>>130
3, 4, or 5 but even fatter would be *perfect*. These turned out really well!

[mod edit: keep ai in this thread]

Hi, has anyone tried to use AI apps like Midjourney to create content?

This is the result so far using this promp:
(((ssbbw))) woman with straight bright ginger bangs, photo, head to toe picture, pretty, confident, round face, obese, happy, (((very fat))), ssbbw, slinky black evening dress, goth, black boots, legs showing, character sheet, concept art, full body, centered, dynamic pose, facing directly at the viewer, comic style,simple studio dark background, analogous colors, stunningly beautiful curvy half-sheep anime girl, perfect anatomy, huge belly

>>9 (OP)
No, absolutely no one has tried at all

Trying out inpainting

I give you more plump elira, this time on a train

>>207
So cuuuuute, thank you

>>215
Great stuff as usual. Could you link me to what model you use or tell me how to make it myself please?

>>217
yeah sure, its a 60/40 mix of grapefruitjuice-pruned and bbwstep30000. you can find bbwstep30000 in this thread a few posts up form the other elira poster.

>>218
Ah, yes I saw your post on /vtai/ too, forgot I asked in both threads. Thank you!

>>218
This the grapefruit model you're talking about?
https://civitai.com/models/2583/grapefruit-hentai-model

fat squirrel

>>222
I'm so glad someone genned this peanut butter pudger.

Would lamias be possible like Miia?

I wonder if anyone has tried bunny girls

Does anyone have a copy of the BiggerGirls Real model that was on CivitAI a while back? I have a slightly blended version saved, but not the original and it's not on CivitAI anymore (at least, that I can find).

>>225

I just have the biggergirlsv2 and bigger girls golden.

Though, tbh, I've had better luck popping the image intp clip studio paint, dropping a nougat colored blob on the belly, and running it through with any model with a low noise diffusion. 0.5 is plenty with the "fat" or "plump" tags.

The problem with the big girl models is that they just remove all the skinnies from the dataset. I mean, that's all well and good, but it trims a good amount of background ambient data out of the set. It's why a lot of these BBW AI tend to look the same.

>>226

This was done with the dalcefo painting model. https://civitai.com/models/5396/dalcefopainting

Started a general prompt with some skinny bitch, and added paint to it (not even good looking paint, just solid colors) and added (fat:1.4), (big belly:1.5), yadda yadda yadda.

Additionally the BigGirls models are all based on Anytingv3, which at this point is several months old. A lot of advancements have been made in the last month alone with quality.

>>218

anyone know where to find bbwstep 30k? I searched this thread for elira step 30k and 30000 and didn't find it....

Made another one

https://bbw-chan.nl/bbwdraw/res/137349.html#141927
I am trying the guide for dummies but I ran into a problem:
It says that Python could not be found. I have python installed. Do I have to move it somewhere?
I also received a message of something being on the wrong path but i cant access that message anymore.
What do I do?

>>225
I had a copy on my pc. It isn't a great model.
https://anonfiles.com/r9J9eaf6zb/biggerGirls_irl_ckpt

>>228
I found this merge labelled as 30k step
https://civitai.com/models/11141/biggergirls-anythingv4-buns-and-thighs

>>231

Cheers, thanks! Definitely agree the Real model isn't great, but for clothed, photorealistic BBWs I find it much better than ForTheMasses. I usually do inpainting manipulations, but I was able to tease out some decent txt2img shots with a BiggerGirls Real 80 + BiggerGirlsv1.2 20 mix. Wanted to play around with it and some Lora's, but obviously couldn't find it anymore.

>>200
She actually has a couple of loras, so i'll see if I can make a few later tonight

>>231
Nice of that guy to upload it, saves me from making an account
>>228
https://mega.nz/folder/WFk13apa#WvCEO8IoqgSaeoKQILeIqg
>>230
I'm only guessing because I'm not on windows, but did you enable "add to PATH" when you installed it? And is your venv set up properly? If the venv is the problem you can delete the venv folder and it'll make a new one on run

>>233
>>235
Could you catbox the third image of both of these posts please? I don't think either made it to vt.

>>232
And the fingers don't look crooked, would be possible to photoshop bbw models with bigger massive bodies?

I like these

Could someone credible by BBW-Chan standards post the Bigger Girls model? I’m not going to make an account for a single download.

>>239
>>231
Ah, it appears someone has posted one. Thank you! Has anyone tested it too?

Since this morning, my images are all black, have you ever had the same problem?

>>241
Ok, it's good, it's my ckpt file that has become defective

I have a mighty need for fat tomboys

A couple of attempts at using 30k step with the hand poser

>>244
Would it be fine to also ask for one of those IRL BiggerGirls models?

1-4 is 30k step, 5-6 is bg2. bg2 always has an issue where the bellies can't go beyond a certain size without inpainting.

>>248
I misspelt this post. It’s supposed to be “(an anonfiles of) one of those non-IRL BiggerGirls models.”

naked apron is hot

another onsen Elira

okay last 2 for tonight

>>256
Love those bellies!

Let's see if you guys accept these B)

Let's see if you guys accept these B)

sorry for the double post, here's some more

>>261
How to delete non-realistic shapes and fat?

Has anyone tried to generate Filia from Skullgirls? I'd like to know what tags to use on her. The hardest thing to nail is obviously the Futakuchi-onna aspect of her.

ITTThe least Asian-looking Asian girls ever

>>227
dalcefo doesn't have his stuff up on civitai anymore, you can download it for free on his kofi https://ko-fi com/dalcefo_artworks/shop

I haven't been having much luck inpainting with dalcefopainting, but it's been going okay with a dalcefo (0.7) + biggergirlsv2 (0.3) mix honestly

pretty inexperienced user here, managed to come up with this and I'm pretty happy with it - though I was wondering if anyone would be willing to uprez / refine it a bit, keeping mostly intact

>>262
These are pretty good, especially the bar one. >>265
Asian-americans exist

>>267
Had a go at it

>>269
came out looking pretty great imo, thanks a bunch

also do you mind me asking what programs you used? doesn't have to be super in depth or anything, just curious
- I only used nkmd Stable Dif UI and bigger girls to start with originally

I'm using a 60/40 merge of LucidMix and BBWstep. I used a depth controlnet for revising your picture to try to keep it as similar as possible. For the pictures I posted earlier, I use the same model, but the base images are from NovelAI. I basically force the furry model to generate humans, which tends to result in a lot more diverse and interesting poses, imo

>>271
thanks for the insight, you've been very helpful - now I have a few areas I can look into further to improve on the generations, much appreciated

>>197
What models and settings did you use to get stocking? I've been trying for a while now but she always comes out looking really bad.

>>236
yeah sure https://files.catbox.moe/4e6lec.png https://files.catbox.moe/1vxxax.png
>>247
That looks super interesting, what extension is that?

>>274
i wonder if there's any way to tell the AI to not add 3 million skin folds in places they shouldn't be. other than that, stellar results anon

>>274
Thank you.

>>274
>That looks super interesting, what extension is that?
Two layers of controlnet. Add multi-controlnet in settings->controlnet in the ui.
Open pose editor for the pose, depth library for hand images
https://github.com/fkunn1326/openpose-editor
https://github.com/jexom/sd-webui-depth-lib

>>146235
What looks alien about them? The faces, body proportions, shading/style or all? I've been trying to achieve bigger but realistic proportions

Some mooms.
Also struggling to get any kind of serious width from the front view. How are you guys doing that?

Hi, I've got a question (please don't bonk me), which programm/software do you need to generate these images? I dowloaded a "ckpt" file called "biggergirlsAnythingv4_bbwStep30000.ckpt" which is the "model" used to make/generate those plump/fat Elira images, but Im not sure of what to do with it. Is there a tutorial for this?

>>251
These are the Elira images Im talking about

>>207
MORE

>>273
Stocking was generated with this LoRA:
https://civitai.com/models/19416/artraccoonee-stocking-anarchy-panty-and-stocking-with-garterbelt

The model is my own. But it's basically a mix of biggergirls + my own model + grapefruit + abyssorange + kenshi

Prompts:
<lora:ArtraccooneeStockingAnarchy_ArtraccooneeStocking:0.6> 1girl, stocking_anarchy, black_dress, aqua eyes, purple hair, colored inner hair, pink hair, choker, bangs, blunt_bangs, bow, cleavage, dress, hair_between_eyes, halo, looking_at_viewer, smile, solo, stocking_\(psg\), halo,lolita_fashion, long_sleeves, striped, striped_legwear, (plump:1.4), (fat:1.4), big belly, round face, (obese:1.2), masterpiece, high quality, highres

Negative prompts:
painting by bad-artist, easynegative, lowres, bad anatomy, bad hands, text, missing fingers, extra digit, fewer digits, cropped, (worst quality, low quality:1.4), normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name

Upscaling and inpainting to fix any oddities with hands/faces if necessary

>>275
Glad you like it
The skin creases & folds are pretty easy to prompt or inpaint out. But I like them, so unless they're particularly bad I don't bother
>>279
Those are some NICE mooms. Love 1,2, & 5.
> struggling to get any kind of serious width from the front view. How are you guys doing that?
There's a few things. First is prompting- all the mooms of this post were done on unmixed bbw30k with (wide hips, thick thighs, fat legs, fat hips, fat thighs:1.1). Wider aspect ratios would also make her wider because SD will look to fill the empty space with more Mumei. If you really wanted to you could also just stretch her hips in photoshop or something and as long as it looks vaguely humanoid, you can use it in inpaint or img2img (nice for sequences, too). Post a prompt or catboxed image so we can see your gen info

>>280
There's a few guides at the beginning of the thread. The program is AUTOMATIC1111 stable diffusion webui, and >>141927 and >>9 (OP) will get you started

>>284
My images didn't post, very cool

>>285
Could you catbox 1, 4 and 6 please?

>>285
excuse me for the silly question but what model was this? I desperately want results like this and have a vae, knowlage of how to use loras, and goldenchubbygirls aswell as some other models.

the massive hips look AMAZING!

>>280
As someone else has said, follow the guides in the OP to get you set up (the novel ai speedrun will get you to download the basic software and a very useful vae file)
If you'd like to generate images like those eliras you'll need my model which is in the "/vtai/ resources: model mixes" link in the OP, scroll down the list till you find "BBWgrape" and that's the one I used. Just note it's specifically tailored to generate vtubers hence why I've been making a bunch of eliras

Interesting artstyle. I got it with 65% 30kstep and 35% 2.5d anime with the blessed vae (saturation between anything v3 and orange mix vaes). There are probably other models that do it better.

Experimenting with a more "cartoony" style.

Mixed results.

>>280
Could you post it?

>>286
https://files.catbox.moe/s27btx.png https://files.catbox.moe/z2xtly.png https://files.catbox.moe/7f3nv6.png
>>287
https://mega.nz/folder/WFk13apa#WvCEO8IoqgSaeoKQILeIqg
Most of my stuff is here including model & short instructions

The poor sketch quality makes it easier to overlook errors.

I gotta hear elevenlabs feederism audio

What's the best model for generating real fatties?

What website do I need to use for real life ai generated images

>>299
>>300
>>295
>>260
>>261
>>262
https://bbw-chan.nl/bbwalt/res/21273.html
https://bbw-chan.nl/gen/res/26545.html
>>298
https://bbw-chan.nl/gen/res/25890.html

>>300
https://genshin.hoyoverse.com

>>274
If you dont mind could you share the BBW30K lora used on those Kronni's please? also does it just enable you to prompt the BBW30k shape on an unmerged model? if so thats really useful.

Can someone help me figure out what is causing the colors to come out like this? While it's generating the colors look fine but when it finishes up it comes out looks crusty and oversaturated. Using the Hasanblend 1.5 model

bbw jean booty

>>297
Without the sketch and pencil prompts.

>>306
Could you please catbox the 2nd and 3rd images?

foxx

>>277
are there listed here?

https://raw.githubusercontent.com/AUTOMATIC1111/stable-diffusion-webui-extensions/master/index.json

I only know how to install extension via the built in web-ui for installing extensions and there are lots of extensions mentioning control net and depth... and even some mentioning open pose's but I am not sure which ones you mean specifically? (the names don't match up with the githubs you linked precisely....)

>>308
Cute!!! What models/lora/prompts did you use?

>>307
https://files.catbox.moe/5956vx.png
https://files.catbox.moe/3gc46w.png
There was a canny and open pose controlnet used for making them. The model was >>289
Cropping the hips/legs off the pose made most of the outputs a bit weird.

>>309
'OpenPose Editor' is in the master list. There are a lot of extensions/tools that can do the same thing.
'sd-webui-controlnet' is also in the master list. Models: https://huggingface.co/webui/ControlNet-modules-safetensors/tree/main
The depth library needs to be installed from url. 'Extenstions'->'Install from URL' in the automatic1111 webui.

Fta gardevoir

Freshman 100

You can pry my "gigantic breasts" tag out from my cold dead hands

Looks nice, but low quality and full of errors.

>>311
Thank you.

>>313

And +300

Does anyone know how to make an immobile girl in Stable diffusion?

>>303
> does the lora just enable you to prompt the BBW30k shape on an unmerged model
That was the plan, but somewhere along the line I fucked up and so it only really does thighs. Really nice thighs, but still. Fats using it look disgusting. If you still feel like playing with it I put it in the mega with everything else https://mega.nz/folder/WFk13apa#WvCEO8IoqgSaeoKQILeIqg
>>318
Download the biggergirls model off civitai and prompt immobile & hyper fat

>>317
and add more plz. want see how far she can go

>>304

you're missing whatever VAE that model uses.

>>294

oh hey! thanks for posting those two of mine, glad people like them! The hinata ones, I mean.

>>319
yeah honestly I've mostly been prompting bottom heavy and mildly fat at best lately so this is perfect, cheers man.

W I D E nina

>>325
one more

To the anon that's generating all of these vtubers, do you think it would be possible to give Juniper Actias a go?

>>327
sorry anon, I've tried but none of my models seem capable of generating her, even the allvtubers lora doesn't seem to work. you could try asking on /vtai/ though.

>>230
>>235
So I started all over again and enabled PATH when installing python.
I still get the red text that says I dont have python installed.
Apart from that I got a text saying "Installing toch and torchvision" and my laptop started going slow. 10 minutes later I got this yellow text and then this green text.
What do I do now?

To bbw-step anon.
I given that you based your model off of Anything v4.5 I was able to do a merge and come up with something that is much better at backgrounds. I am currently looking at other merges to see if it can be improved further, but the following pictures were all generated using the half version of the custom merge. Note the images have not been inpainted, so there is some strangeness that hasn't been cleaned up. All of these have either had hires fix applied, or have been upscaled using Ultimate SD to 1024x1024

>>330
Also, I noticed that sizes tended to drift towards the smaller end, but it should still be possible to generate pre-merge sizes with very little effort.

>>330
I did two more merges.
The first image in the original post is being regened in the first and second images of this post. Same prompt, same seed, same settings.
First image is the merge, merged with Cocoa. The second image is the cocoa merge, merged with tea.

>>332
Prompt for the three I mentioned are the following:
(extremely detailed CG unity 8k wallpaper, masterpiece, best quality, ultra-detailed, best illustration, best shadow, an extremely delicate and beautiful)
<lora:allVtubersLora_hll31:1>, ouro kronii, hololive, headband, crop top, jeans, muffin top, belly tucked into jeans, too tight jeans, wide hips, fat, huge belly, table, hands on belly, stuffed, helpless, sitting down, too full to move, wavy mouth, looking at viewer
Negative Prompt:
(worst quality, low quality:1.4), (depth of field, bokeh, blurry:1.4),(monochrome, greyscale, monochromatic) ,(multiple views, comic, reference sheet), motion lines, error, bad anatomy, by (bad-artist:0.6), midriff, (fat legs:1.1), sketch, 3d, flat color, extra digits:1.2, fewer digits, deformed, extra limbs, fewer limbs, collar, brooch, necktie, bowtie, necklace, bad-image-v2-39000, EasyNegative, ng_deepnegative_v1_75t

Tried my hand at this and got some decent stuff

>>329
> I got a text saying "Installing toch and torchvision" and my laptop started going slow
thats normal, torch & torchvision are several GB large and take a while to dl and install
> I got this yellow text and then this green text
you can disregard the yellow text- it says that it failed downloading torch, but in the fourth screenshot it says torch installed just fine, so it probably just pulled from a different source. And as for the green text you can update pip if you want but it doesn't really make a difference, so it's up to you

The fourth pic would suggest that everything installed fine and it just isn't picking up your graphics card. By default the webui doesn't work with AMD cards on windows so if your laptop is AMD then you have to follow the guides on this page https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Install-and-Run-on-AMD-GPUs or the AMD install guides at the top of this thread >>9 (OP) . If you're on intel graphics, it doesn't work with intel graphics, so you need to enable cpu mode to get it running. and if youre on nvidia, then i don't have a clue. it should recognize your gpu out of the box. reinstall drivers?

>>325
Could you catbox that please?

>>336
this isn't AI shit it's just traced

Here it is. The merged bbw models. Pic related is a comparison between all of the combinations I made using a prompt that some might use with this model. Kuki needs more work, the model I merged with needs less aggressive weighting on certain blocks of the unet so that bbw-step isn't diminished too much.

https://gofile.io/d/AzRqHS
(All of the models here in the above link need to be pruned. )


Some general notes regarding the model:
The model (and the original bbw_step model that the merge is based on) is based on Anything V4.5, which means that this should be compatible with most LoRAs and embeddings. If you want to get backgrounds that are even more detailed, using the token "detailed background" will activate dpep3 much more aggressively. As for negative prompts, go for whatever you would use for an Anything v4.5 or AOM2 based model.
Some notes regarding the Cocoa and Tea Merges:
These two have some oddities introduced from Cocoa which causes the masterpiece keyword to create some weird results, although not consistently, so experiment with it in your prompt and with it not in your prompt. I have had good results with and without it. Additionally, make sure your clip skip is set to 2, otherwise you might encounter some oddities.


Recipe for the merged model
First, merge dpep3-chillout into bbw_step_30000k (when I say this, you should make bbw_step model A, and dpep model B)
(find it here: https://huggingface.co/closertodeath/ctdmixes/blob/main/dpep3-chillout.safetensors )
with the following Merge Block Weights (you cannot use the default checkpoint merger, use supermerger, or another extension that can do MBW)
1,0.9,0.7,0.5,0.3,0.1,0.4,0.4,1,1,1,1,0,0,0,0,0,0,0,0.1,0.3,0.5,0.7,0.9,1
(You can start from the half version in the above link if you want, the fp16 and normal version did not have any hash differences)
Then merge the previous model you just made, I would call it bbw-dpep3-chillout, with Cocoa at the following block weights:
0.15,0.15,0.15,0.15,0.15,0.31,0.3,0.3,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15,0.15
(NOTE: FOR ANY MERGE INVOLVING SPECIFICALLY THIS MERGED MODEL, ONLY DO MERGES WITH THE FULL SIZED MODEL AS THE HALF SIZED ONE HAS A DIFFERENT HASH FROM THE FULL SIZED ONE)
Then merge the previous model you just made, I would call it bbw-dp3c-Cocoa, with Tea at the following block weights:
0.35,0,0,0,0,0,0.1,0.1,0.1,0,0,0,0,0,0,0,0,0,0,0,0,0.35,0.35,0.35,0.35

(You can find Cocoa and Tea here: https://huggingface.co/andite/desserts/tree/main Cocoa is most likely a mix of YohanMix and Blossom-extract, but I have very little information on Tea )

Anyways, let me know what you think, I feel like the merge managed to really fix a lot of the issues the base model had. My favorite is either cocoa or tea, I haven't done enough prompting with either to really come to a conclusion.

Alright fellas, as we approach the bumplock for the 5th thread, I gotta say: damn. This is one of the most successful repeated threads ever. Are we all ok with just archiving the bumplocked ones and making new ones, or is it time to make a cyclical thread again? These threads move so fast it's a bit difficult to get all the information in one place, a cyclical thread that I start where the OP can be edited as we go might be easier than having 5 archived threads of information to sift through. Just a thought.

>>335
I looked it up and realised I use Intel Graphics (I should have looked that before, Im a fucking moron)
So how do I enable cpu mode?

>>340
That sounds like a good plan.

>>9 (OP)
genuinely didn't know animation generators were this advanced already.
I expect good things!

Does anyone know on Stable diffusion how write a specific character prompt with out a checkpoint/model like if I wanted to make Yor from Spy x Family

It's nice to see all of these people come together and bond over AI generated images in harmony.
I know it's a fucking meme phrase at this point: but I am actually proud of this community.

>>339
I gotta say I really appreciate your help in making this scuffed model usable. How did you figure out the weights to use with supermerger? Do those specific weights control mostly backgrounds or something?
>>340
welcome back. I was wondering where you went. And yeah, that does sound like a good idea, as long as we link back to these archived threads
>>341
open your webui-user.bat and add --skip-torch-cuda-check to COMMANDLINE_ARGS. Idk if theres anything else to use cause I don't use cpu mode, there's a guide for it in the op

This whole multi-character openpose stuff is wild my dudes I'm really kinda of floored how fast this tech moved. For the average layman, to absorb the last 6 months of updates/changes/extensions/controlnet/lora/textual inversion/inpainting/etc must be RIDICULOUS if you haven't been following it the whole time.

Can we start collecting some vital links/resources for the cyclical thread? I don't just want to post the /vtai/ blob of text again, I want something more...us.

>>344
depends on the character, and ranges from 'annoying' to 'basically impossible'. Easy way to go about it is to just pull up gelbooru or danbooru and copy the character tags, and if your character is popular (which would mean they have more images on the boorus for SD to be trained on) then they'll probably show up. But it's only for anime characters, and it's pretty hit or miss even if your character is popular. and if they're not, you can forget about it (RIP buffpup & juniper anon). Loras can help with making characters but if you don't want to use them for some reason or if one doesnt exist then you're SOL. Yor has a nice one on civitai

>>346
>>348
Could you catbox that IRyS and that first Yor (36776) please?

Very fat Lumine genshin impact, ripped letter black jeans, huge ass

Thick girl big boobs

is there any way to get bbw step 3000 to generate dark skin? i've put in all the tags related and it doesn't really do much

>>350
>>351
I don't think you understand what this thread is

>>350
>>351
ESL bois or pre-teens have found the thread.

>>347
>>347
I'd keep the guides for newbies and the installation guide,

add links to the BBW3000 mega, the new tea/coffee merges and any other merges made (perhaps even make own rentry for models),

a guide to prompting bbw stuff in particular,

an updated guide for generating prompts, upscaling and inpainting etc - basically walk a newbie through how to generate a high quality image.

Perhaps a personalised FAQ, eg what bbw models exist, how to prompt etc.

a general FAQ like why we use catbox etc.

There's just a few examples of stuff that might be useful but I'm sure there's more I haven't thought of.

Does anyone know any models/loras for both all bleach girls or all total drama girls?

>>105
>>106

Damn, I need to try this. Liara in the style of these super thick, bottom heavy hourglass ones would be fantastic

An attempt at training a big belly lora. Not very good since it ignores most of the trigger words I tried to add and occasionally does an extreme close up of a face or feet. Will need to adjust the tags again and retrain it. My amd gpu caps at 640x512 for training.

The trigger words are supposed to include
>1girl
>solo
>round belly
>flabby belly
>skinny
>fat
>one leg hidden
>both legs visible
>full body
>sitting

Really only responds to "round belly". Square aspect ratio usually gives a front view. Side view more often with a wider aspect ratio. Clip skip 2 is slightly better.
Looks terrible most of the time. Gets okay results if you don't use the lora at full weight and upscale 2x at low denoising with r-ersgan anime6b.

https://anonfiles.com/i1f8M1gazc/blobv3_safetensors

Sample catbox (model is 65% 30kstep, 35% anime2.5d): https://files.catbox.moe/kgnhhe.png

Thanks to LtBarclay's guide in the 4th AI thread, I was able to get the web UI running. Big thanks for creating such a helpful guide!

My AI model was created using biggergirlsv2 70% + barronmix 30% and mixing the result with 3moonREALSkin model at 30%. I was also using misty LoRA at 0.7 for these images.

I've found that very painterly effects could be achieved using DDIM sampling, 8-12 sampling steps, and a CFG between 3 and 6.

So anyone know of a safe place to get ?
animefull-final-pruned
animevae.pt
Sorry if this is a dumb question I googled it and couldnt get any of the downloads to work.

>>358
Another attempt, still not great. This one gets consistent poses starting at 0.40 lora weight. There is a bit less variety after tweaking the training frequencies. Trained at 512x512 max because 640x512 wasn't working this time.

https://anonfiles.com/T720W2g5z5/bellyv4_safetensors

>>361
Thank you for including the tags.

>>349
https://files.catbox.moe/detvw0.png https://files.catbox.moe/wyhb8m.png
>>360
i think they should still be available via torrent in one of the links in the op. Don't remember which one, maybe nai speedrun?
>>355
really good suggestions, especially making a new rentry. Basically everything you need to know is here but it's in bits and pieces over 5 (massive) threads so consolidating everything into a rentry or two would be great
>>352
works fine for me? "dark skin" works as expected. adding emphasis to it or using it in tandem with 'black skin' makes them darker. Although it is a little random, as always https://catbox.moe/c/ord80b

>>363
Thank you!

been testing out the style lora's someone uploaded to /vt/ here's a Nina using the bamboo ale lora for example.

MEGA link for anyone interested: https://mega.nz/folder/sShj1LpS#Zh5ZxGO3uwK4QEaZ2X6z1w

>>365
Yes, artists who hate AI must have loras, based on their arts

>>365
Do you use nai or this can be work with other vae?

>>366
Not like they can do anything about it anyway. Imitation is the purest form of flattery

>>363
oh yeah thanks, realised all i really had to do was emphasize the tag and it turned out fine

I'm starting to think actually making a whole new /bbwai/ board.
Maybe a wild hare here, but think about it. Our own stickied threads for the basics.
A requests thread. A model compendium and review thread. A promptsmithing thread.
Specific threads for specific vtubers. A thread for guides. A thread for sharing wildcards and textual inversions.
A thread for collecting and testing Loras. I mean it just goes on and on.

The reason I bring it up is because, I've been more active modding the past 3 months than the past 3 years because of AI.
Maybe it's time to give it it's own little space.

>>370
Please do make a board for AI.

>>370
I'm not necessarily sure its big enough to warrant an entire board just yet. There's around 60 unique posters on this thread and sure more will be drawn in by an entire board being dedicated to the subject but will it be enough to sustain itself? Its hard to say imo although it would be very cool in theory. I think a decent amount of stuff could simply go into a rentry for this thread eg: guides, models, model reviews, LORA archive etc. but thats just my two cents, it's definitely a cool concept.

>>372
*possibly ignore the part about 60 posters, was going off of the "unique posters" at the bottom of the thread but that doesn't seem to be accurate / consistent across threads.

>>370
the biggest problem I see right now in both the current state (isolated ai threads on different boards) and a single ai board is where should niche things go. As it currently stands, things like loli, furry, etc. get shunned out of the general ai threads to /bbwalt/, but then those threads there say to go back to the ai threads. If a new board is made for ai, the rules for that content need to be made clear so no one complains later. Will the ai board be all inclusive (all ai pics go there) or just /bbwdraw/ level stuff? realistic? Or should there be a general ai board with another one for alt stuff like a /aialt/? I don't know what the real solution is but as long as the different alternatives are considered and then clear rules are made, I'd be good. Also in the current state, I don't know if there are even rules for where ai can be posted.

>>370
Please do this!

>>372 >>373
I agree. Across the whole site there's maybe ~5 active ai threads and they're all slow as hell. This is the most active thread by a long shot and even then, of the 60 unique posters, maybe a a third have posted any images, and even fewer post consistently.

But maybe there should be a board anyway? If you don't care about what a tiny and slow board it would be, it would certainly help keep things organized, since the ai threads are scattered over 3(?) different boards, and people keep coming here for real & voice. And I would bet that people who aren't into ai would still very much appreciate there being an ai containment board and just banning it on the rest of the site, since it's universally despised by anyone who doesn't make or consume it.

>>370
> A requests thread. A model compendium and review thread. A promptsmithing thread.
> Specific threads for specific vtubers. A thread for guides. A thread for sharing wildcards and textual inversions.
> A thread for collecting and testing Loras.

Even if the ai threads were 100x busier this would be excessive. I think it would be best to have a thread for drawn, another for real, another for furry, then loli, then voice, etc. And then guides & loras & models and such should be in the OP of their respective thread, and requests can just go anywhere in their respective thread, maybe onto a request anchor.

>>374
If we don't make a cyclical thread here, in furry, and alt, then the next best option is a giant catch-all board. Regular, SSBBW, Furry, Alt, Text/Chatbots/Voice: if it's AI-generated, it would go there and everyone would just need to get along..lul.

>>376
>But maybe there should be a board anyway? If you don't care about what a tiny and slow board it would be, it would certainly help keep things organized,
The lifeless husk of /preg/ should show we don't really care if a board is slow. And yeah, mostly what I'm thinking with the board is consolidation, just getting everyone who's interested in AI in one place, while at the same time appeasing those who wish it would 'go away' or be shunned to it's own area.

>>365
Thank you, i use bamboo lora, sorry for not uppscale