@LettucePie it's been a component of my work for a few years now in the form of tone transfers, sample generation, etc. as well, the "photos" of the scenes greenhouse is in on the previous record are generated (foxsnax's illustrations on top of them are of course not); what was i supposed to do, learn 3d a render a dozen detailed scenes? build a bunch of detailed sets irl and take photos? nevermind the fact that so many of the places are physically impossible and horribly misshapen, hahaha

as with anything, it is what you do with it that counts. machine learning in art is like cloud gazing, and an enormously powerful tool for artists. anybody generating something and touching it up via inpainting etc is missing the plot totally.

in terms of impact, it is not unlike virtual instruments before it (which displaced session musicians, but find best use with independents simply seeking to realize their ideas), or recorded music before that, which encroached on live music. each innovation makes the previous thing easier for the "average person". and i think that's probably ultimately a good thing, though the system we live in (capitalism) makes it a very challenging thing.

as for the ethics re:style/idea/etc jacking, i agree that there's much to be said about the ethics of model harvesting. but when you go this deep down the rabbit hole in terms of model checkpoint merging, the chances of you generating something that any human would create on purpose is hilariously slim.

tl;dr use it for what it is, not to replace people. the computer's malformed dreams are strangely inspiring.