Hey, I keep seeing your models of Madison, and each seems to need some improvement. I thought I could offer you some advice if it's not unwanted.

Firstly, you should use clip skip 2 when training on anime-like characters.

Secondly, you could try using a lower dimension, like 16 or 8, so that the model doesn't retain unwanted information.

Thirdly, you should try using more than one image.

Finally, when training, check the shuffle captions box; it could help improve your models.