This is a GLora trained from koikatsu screenshots I captured in 4 scenes (360 screenshots). I chose GLora because it doesn't overfit too much. I also avoided soles in the training set.

Please do not put "???????" in the prompt, or it will create bad results (????? towards up).

It works for most anime checkpoints, but works better in 2.5D models than line arts. If used in line art models, it will look like koikatsu scene. Sometimes the ????????? won't show because the training image does not show them often (obscured by foot). Also it's not working well with shoes (but possible).

Recommended resolution: 512x960 or 512x768 and I used "Restart" as sampler.

In the future I will try to train it using more screenshots and diversify the training set.

I hope you enjoy it :)