Overview
A LoRA to create (very) thick female bodies. This model cannot produce X or XXX on its own and honestly most R would be considered PG-13.
The model is pretty straight forward. Use the trigger thicc early in your prompt. Further descriptions to the body, hips, thighs, framing and focus can help a lot to emphasize the effect.
Values over 1.0 may produce heavy artifacts. Decrease strength if you encounter artifacts. Most samples use a strength of 0.9. Can be combined with Macromastia (Clothed) or even the naked version.
For more have a look at the example images/videos.
LTXV 2.3 - V1
Coming late July...
Krea 2 Turbo - V2
Coming late July...
Krea 2 Turbo - V1
Use the WAN 2.1 VAE for more realistic outcomes. Use the filter bypass LoRA for better results.
Training model: Krea 2 Turbo + Training Adapter
LoRA strength: 0.5-1.0
Sampler: euler / euler a
Steps: 8+
VAE: WAN 2.1
A photo of a thicc woman standing in a dressing room, wearing a long red dress. Her curvy body, wide hips and thick thighs stretch her dress thin.WAN 2.2 T2V - V1
I mainly use this to improve the results of Thiccification but it can be used solo.
Training model: WAN 2.2 T2V
LoRA strength: 0.3-1.0
Sampler: euler / euler a
Use TripleKSampler with 1/4/4 steps or no lightning LoRAs at all for quality.
Use thigh/butt physics LoRAs to improve jiggle and motion.
A thicc plus-sized woman standing in a bathroom, wearing a sweater, high waist jeans, in a bathroom, she has a chubby body with a slightly rounded belly, wide hips and thick thighs, medium shot,Flux.2 [klein] 9B - V1
Training model: Flux.2 [klein] Base 9B
LoRA strength: 0.3-1.0
Sampler: euler / res_2s
Steps: 4+
CFG: 1.0 (up to 5.0 with base model)
An amateur photo of a thicc 21-year-old woman standing in a well lit bathroom, wearing underwear, she has a fit body with narrow waist, wide hips and enormous thighs, she leans forward onto the sink; big butt upwards,Z-Image Turbo - V1
Not compatible with Z Image Base.
Training model: Z Image De-Turbo
LoRA strength: 0.3-1.0
Sampler: euler / res_2s
Steps: 8+
CFG: 1.0
Prompt example:
thicc, A photo of a young woman standing in an office holding a presentation, wearing a tight elegant black dress hugging her generous curves, she has a narrow waist, extremely wide hips and thick thighs,Description
FAQ
Comments (15)
When training, how do you keep the face/identity from training?
My models definitely influence the face in some way, but I'm trying to get better at is. Key is either blocking the heads/faces, which leads to other issues. The best way is to have a very large varied dataset and use a timestep type that focuses on larger shapes/higher noise, instead of details (like shift, instead of sigmoid).
Versions for Qwen Image and Qwen Image Edit please!
Sorry, Qwen is currently not on the roadmap 😅
The body shape is incredible, however it seems to influence the face and lighting quite a bit. Hope to see a version 2 with an expanded data set!
Yeah, it's only about 80 images. I only trained it because Krea 2 trains fast and I had it lying around. Will definitely improve on this one!
When Krea 2 comes out for https://github.com/shootthesound/comfyUI-Realtime-Lora, you may be able to disable the layers that influence the face, for a workaround until I've trained a V2.
Good to hear! Thanks for the link - I haven't seen this project before but it looks very useful
@cupra do you mind sharing the training images and info ? What did you use to train it?
@HiProfileArt hi, I can't share my datasets, but the training parameters and setup. It's a bit annoying to share it here, but you can join my Discord, if you like.
Illustrious version?
Not needed. Plenty of similar LoRAs already there, like Venus Body.
Works with I2V. Love it!