this post was submitted on 16 Aug 2024
91 points (89.6% liked)
Showerthoughts
29172 readers
212 users here now
A "Showerthought" is a simple term used to describe the thoughts that pop into your head while you're doing everyday things like taking a shower, driving, or just daydreaming. The best ones are thoughts that many people can relate to and they find something funny or interesting in regular stuff.
Rules
- All posts must be showerthoughts
- The entire showerthought must be in the title
- Posts must be original/unique
- Be good to others - no bigotry - including racism, sexism, ableism, homophobia, transphobia, or xenophobia
- Adhere to Lemmy's Code of Conduct
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Arguably, AI struggles with hands because the creators of the training data struggled with hands.
AI sucks at hands precisely because humans suck at hands.
It's also because generative AI relies on statistics so an approximation of the number of fingers is used. In that sense 4 and 6 are close enough to 5.
A third reason is that not all fingers are always visible in pictures and drawings, depending on the angle and position. If you try right now to make a 'c' with your hand, you could only see three fingers. This is relevant as while artists know that hands have five fingers, generative models do not 'know' what a hand or a finger are.
Just to explain more of this "approximate amount of fingers": you might have seen this in the wild, imagine the AI starting out with white noise, and then slowly brushing over different parts to add different things the model would expect. There's gonna be areas where the AI expects "heads" and "limbs" and finally "fingers". But it's more like a texture. Those AIs have no direct concept of what the right amounts are for animals and humans. That's why they paint "finger texture" in about the right spot and move on, and of course that's gonna look weird 50% of the time.
There's actually specific techniques modern AIs use to make sure any human or animal has the right amount of limbs and stuff because the AI on it's own would never learn how to do this well otherwise. With text it's really similar btw., although at least you can see the AI nailing the type of font before failing at coherent text. New AIs do a separate pass to recognize specific text in the prompt and then manually add that after the base sampling process iirc.
AIs are so interesting on many levels and the more you learn about them, the more respect you gain for artists and how much work goes into a piece of art. And it also really puts in perspective that (my opinion) AI will never completely replace artists. Our expertise on arts is just too strong so for the foreseeable future AIs will just be a solid helper to every artist at most.
Not to mention animators have traditionally gone with 3 fingers and a thumb, making the AI have to work even harder.