I made this LoRA a little over a month ago, and honestly, it's sucks. The huge issue is with Magus, since on a ton of the images Magus's "head" is peeking out from the edge of the frame, which tricks the LoRA into thinking the "head" is separate from the tail ;/

I think that right now it's possible to put together a better dataset, exclude images where the "head" is separated from the tail, and still maintain flexibility in clothing but but I'm too lazy