24
TIL a lot of people think AI art tools just copy existing images
I saw a thread where folks were arguing that tools like Midjourney just stitch together bits of other pictures. That's not how diffusion models work at all, they learn patterns to create new stuff from noise. Has anyone else had to explain this to a friend?
3 comments
Log in to join the discussion
Log In3 Comments
morgan.jason14d ago
It's more like they learn a visual language, but the training data is still a huge part of that. Hard to call it totally new.
7
jason_stone5914d ago
Man, this reminds me of when my dad saw a picture I made with one of those tools. He swore up and down it had to be a photo of a real place I found online. I spent like twenty minutes trying to explain the noise-to-image process and he just kept nodding and saying "uh huh, sure." Some ideas just get stuck in people's heads, you know? They see a cool picture and their brain jumps to the simplest explanation, even if it's wrong.
6