Hands can have so many variations of position in 3d space that it makes determining what is within realistic thresholds difficult on a 2d canvas.
Breasts are easy, similar shape, pattern, and position in relation to the body. Lots of training data. Can still get stray nipples and partial or multiple breasts.
Faces are harder, more acceptable variations but still limited configurations and fairly limited positionally. Faces looking up are really difficult due to lack of training data.
Arms and legs are harder again due to having to render multiples and having enormous positional variation.
Hands, lots of moving bits, 20 finger segments that can sit in thousands of configurations, but not all of them are realistic, AI is still learning the proverbial limits.
![Care-free [SD]](https://lemmy.world/pictrs/image/8e66f1c5-fe69-4500-9c6b-c4d1a18bbbb2.jpeg){kind=link}
Hands can have so many variations of position in 3d space that it makes determining what is within realistic thresholds difficult on a 2d canvas.
Breasts are easy, similar shape, pattern, and position in relation to the body. Lots of training data. Can still get stray nipples and partial or multiple breasts.
Faces are harder, more acceptable variations but still limited configurations and fairly limited positionally. Faces looking up are really difficult due to lack of training data.
Arms and legs are harder again due to having to render multiples and having enormous positional variation.
Hands, lots of moving bits, 20 finger segments that can sit in thousands of configurations, but not all of them are realistic, AI is still learning the proverbial limits.
Also AI can’t count, only guess.