/vt/ - Virtual Youtubers

>>73028780
As for hands, the main issue is they come in too many shapes and the fingers are too thin for the ai to properly perceive/draw them. The SD1.5 latent was 64x64. In that size, even a single pixel would be too big to represent a finger. That is why SDXL improved hands with the biggest problem right now being it is making too many of them like in the clara pic.

>>73028973
Wouldn't that mean the clara pic should have had the hat and ponytail still on the right side of her head even if she turned around because that's where the AI knows it should be positioned?