Update: I keep seeing people think AI art is just a filter
It's NOT just pressing a button, the models are learning from millions of images to create something new. I read a paper from Stanford showing how diffusion models build pictures pixel by pixel over many steps. Has anyone else tried explaining this to a friend who didn't get it?
That Stanford paper is interesting but I read a different one from MIT last year... it showed how the training data itself is the real work. The models are just mixing what they've seen, they don't understand what a "new" thing is. It feels more like a really advanced collage tool than someone learning a skill.