First, hands are just super complex in the first place; it’s why they’re also hard for humans to draw. The geometry is super complex—there are tons of bones and tendons, they move in a million different ways, and they look completely different at different angles.
It also has to do with how AI works. It has to extrapolate information from a vast data set, and with all that hand complexity we were just talking about, the AI can’t get “consensus” on what hands are supposed to look like. So, it does its best and it averages out all the different possibilities. That works with a palm tree—its not important if that has 20 fronds or 25—but it has no way of knowing that it IS important for there to be 4 fingers and a thumb on each hand.
![Having fun on vacation [album]](https://lemmynsfw.com/pictrs/image/16705012-ebe8-4404-8613-4c54b1744da0.png){kind=link}
A couple reasons, IMO.
First, hands are just super complex in the first place; it’s why they’re also hard for humans to draw. The geometry is super complex—there are tons of bones and tendons, they move in a million different ways, and they look completely different at different angles.
It also has to do with how AI works. It has to extrapolate information from a vast data set, and with all that hand complexity we were just talking about, the AI can’t get “consensus” on what hands are supposed to look like. So, it does its best and it averages out all the different possibilities. That works with a palm tree—its not important if that has 20 fronds or 25—but it has no way of knowing that it IS important for there to be 4 fingers and a thumb on each hand.