This one was really fkn hard to make. The dataset could be better and bigger with better tagging but I've done all I want to do with it. There's TWO versions (AND YOU'LL NEED BOTH) mostly because I don't know how to combine LoRA's. The anime one is a little more stylized and the proportions will be less realistic. I like to mix the LoRA's together. I find it only really works when you use more than one. This is basically a WIP that I will never finish. Thought I would release it since it kind of works.Trained on 1000 ish images with mediocre tagging on 80% of them. Putting 'lying on person' seems to help.

Reverse Headscissor

This one was trained on anime images. Again, use both versions in your generations

Files with SHA256: 852009e4a5da4c63971757a821aba515ac6fa1fb1ecf57979d85c59505a8efc4