>>35126 how do you manage to achieve corsets on fat ppl?, i just cant get it to work

>>35126 >>35127 Oh my god this needs to be a series, especially the woman in blue and the one sitting down in white

>>35142 its been hit and miss on corsets - but describing the corset / clothes in detail and being more simple about the subject's obesity seems to make the generator think harder about the clothing. completely rewriting prompts can drastically change the result of very similar descriptions - I dont save the prompts because I'm looking for variety and want different people, tapping out a fresh description does that. ^^ On the blonde in the corset image I do remember adding a "Emphasis on abnormally blubbery fat arms with creases... " etc etc etc at the very end.

These are pretty sick ngl.

... A few more fat fair girls.

this batch with a fatter version of the girl on break. Was going for super fat but believably. Anyone ever able to get "believably fat" but with stomach blubber that hangs below the knees when standing? ... or even sagging onto the ground - inside a pair of leggings or something? At a certain point ImageFx just turns out run-of-the-mill ai blobs that are huge and less real

>>35161 Iveen trying comfyui and i dont seem to be able to make remarkably fat bitches in it, tried Modesl like cyberrealism pony, pony realism, illustrous, and now realdream tried different loras, like: https://civitai.com/models/675260/extreme-obesity-ssbbw-generator-for-females and https://civitai.com/models/461046/immobile-extreme-fat-pony-xl?tags=3915 fuk am i doing wrong? i dont even want blobs, just like, 600lbrs or something

>>35162 we need to train a model (lora), theres a decent one on Flux but Flux is not realistic enough for me. I have trouble gathering a dataset for WAN 2.2 lora's because models are so lazy and don't use good photo quality.

>>35165 unrelated, but I can tell that you arent filtering masculine features. In the negative prompt, include anything masculine to fix this. Also, you should try https://civitai.com/models/277058?modelVersionId=1522905 I have not experimented much with ssbbws, they aren't my thing, but they are possible, although the quality and stability drops the larger you are trying to make a woman. Provided are examples of how sixe correlates to realism

Been playing around with WAN

>>35225 Got somewhat decent results with https://civitai.com/models/2664/bigger-girls-models?modelVersionId=56154 though this has adetailer and the face is still fucked up. the model is not bad but i feel you have to be more specific than usual as to not get glitches. I still dont know how to do the face-swap thing nor complex comfyui setups so might be screwed for the time being. Last one was made with epicrealism, its k

>>35258 >I still dont know how to do the face-swap thing if you dont feel like setting up the workflow, there are some to download. try out: https://github.com/amao2001/ganloss-latent-space/blob/main/workflow/2025-02-10%20ace%20reference%20editing.png

sharing prompt cuz gatekeeping its just dumb 373558 natural colors, daytime, sunrays, before after comparison full body shots of impossibly weight gain in 5 years of female armenian college student, tanned nose, she wears natural makeup, eyeliner wing, fake eyelashes, blonde ponytail, brown eyes, freckles, pierced navel, tiny black twisted bandeau bikini, sideview at her kitchen, with lots of enormous jars with weight gain and body mass suplements and shakes on top of the kitchen counter. In the before picture she is extremely skinny and fit, held a tiny white marker board with the written sentence "fat scale challenge" with food doodles, softly hand on her waist, happy. in the after picture she has expanded into a gigantically wide mountain taking up the entire room as her waves of fat spill outward in all directions, is impossibly obese, she is rapidly growing fatter, she is beyond the fattest woman in the universe, magic weight gain has caused her to swell to impossible fatness, every limb is extremely swollen with fat, particular emphasis on her gigantic fluffy body, obese cheeks, holds a big funnel connected to a thick garden hose, hand on her hair, proud of her gain, shy smile. sideways. at the back festive letters balloons "500!",

383619 natural colors, daytime, sunrays, blury background, before after comparison full body photos of impossibly weight gain in 5 years of female argentinian college student, extremly tanned skin, she wears natural makeup, eyeliner wing, fake eyelashes, blonde ponytail, brown eyes, freckles, neck choker, black twisted bandeau bikini, sideview at her kitchen, with lots of enormous jars with weight gain and body mass suplements and shakes on top of the kitchen counter, tiny white marker board with the written sentence "road to ssbbw" with food doodles at both photos at the same spot. In the before picture she is perky, softly hand on her waist, happy. in the after picture her bikini doesnt fit, her skin is marbelled with skin folds, she has expanded into a gigantically wide mountain of fat and creases spill outward in all directions, is impossibly obese, she is rapidly growing fatter, she is beyond the fattest woman in the universe, magic weight gain has caused her to grow to impossible fatness, every limb is extremely obese, particular emphasis on her fluffy obese arms with creases, her layered fat rolls ressembling Michelin mascot, obese cheeks, happy pleasure. shrugging one shoulder, one hand on her waist, measuring tape, poses sideways, blender jar with tons of butters.

>>35347 "only 500" pathetic

>>35355 Lol but my idea is to make it so its believable.Realistic as thread says xD. im happy you played with that prompt. its a mix of all that has been said/shared here if u read carefully

you can drag png into comfyui for workflow

>>35450 >>35449 >>35448 >>35447 holy hell, what model/workflow do you use?

>>35447 This isn't true I don't think. At least I can't open these images as Comfy workflows. Maybe the chan strips the metadata from the uploaded images? Would definitely _really_ love to see your workflow.

>>35469 I take it back, was able to load it. Great stuff, thanks!

>>35470 Great! Glad to hear it worked im curious to see others results too! Image to Video works great as well

>>25945 Reup please?

Been testing out a new lora I made.

>>36204 I have 2 versions of the lora that I trained on qwen and SDXL. The SDXL one I've been using with Pony Diffusion 2.23, but it seems like it's harder to get naturally realistic elements. It still looks painterly in parts of them. But if I use the qwen lora with the qwen checkpoint, then things start to look much more real.

>>36206 great work mind sharing your lora?

>>36209 Definitely! I'll be dropping it on Civit AI once I know it's good enough. I was trying to parse through what you were doing to make you image and video gens, but instead, would you mind making a consolidated post of your checkpoints and loras?

>>36209 Here ya go https://civitai.com/models/2038661?modelVersionId=2307398

>>36217 thanks man, do you mind sharing your dataset? I can train it for Flux 1 dev (krea) if you want :)

>>36267 Sure, but the dataset isn't the greatest. This was more of a test to see how a large dataset worked with not great captions. The captions would need a lot of cleanup and there are over 1800 images with them. The captions were auto generated for me with a few key words added. I manually cleaned up only a handful of them. https://gofile.io/d/csc6C9 However, I did make a wildcard with a huge number of prompts already in it, so you can just run a long generation and get a nice big variety without having to baby it. https://justpaste.it/8u0iq