Type | |
Stats | 7,010 |
Reviews | |
Published | Apr 25, 2023 |
Base Model | |
Trigger Words | feet101 |
Hash | AutoV2 D9A3C258E1 |
k-Feet Mix (V1)
Suggest to use weights 0.9 or larger, with a trigger word "feet101". Try not to use CFG Scale more than 8~10, I use 7. But some people lowered weights to around 0.5 for anime images, and it seem to work well.
I guess it might work generally when there are feet in images - please try to tweak weights/CFG/etc. to better suit your needs.
This LORA was trained on close-up female feet images, so it may not come out well on images other than close-ups.
This LORA was trained on a limited number of feet images with a limited number of feet poses. So the variations you can expect from this will be quite limited.
The chances of getting a decent foot image is, I would say, around 30% on my settings (by decent, i mean, it has 5 toes and look okay).
Plan to develop more on feet.
Good luck and I'll come back with the next version in the future, if this what people want.