Are we perhaps already seeing the indirect effect of diffusion model art here? When photography was invented, artists pivoted to unrealistic art. Now with text-to-image models getting better and better, even technical drawing prowess for unrealistic art is becoming unimpressive. So artists and creators have to find the interesting aspects of art elsewhere, like in the unusual personality or circumstances of the artist.
I was with you on the first part, but I don’t think this holds up. The sorts of things that diffusion models struggle with—functionality, for instance—are pretty easy to work into an image that still looks good. For example, a system of gears that spans an image, or a rube goldberg machine.
It’s getting trickier to put into words what diffusion model outputs are characterized by, but people can still implicitly tell, most of the time. This is not to say there hasn’t been an impact, ArtStation front page slop is a fairly dead medium nowadays, but I think it will still be quite a while to exhaust the set of artistic skills that are both hard to replicate and visually beautiful.
I was with you on the first part, but I don’t think this holds up. The sorts of things that diffusion models struggle with—functionality, for instance—are pretty easy to work into an image that still looks good. For example, a system of gears that spans an image, or a rube goldberg machine.
It’s getting trickier to put into words what diffusion model outputs are characterized by, but people can still implicitly tell, most of the time. This is not to say there hasn’t been an impact, ArtStation front page slop is a fairly dead medium nowadays, but I think it will still be quite a while to exhaust the set of artistic skills that are both hard to replicate and visually beautiful.