I think the reason we're seeing better AI images is not because the programs are getting better, but the users are getting better at ignoring the chimeras and finding the one and only generated image that is halfway decent (and sometimes it's more along the lines of "obvious tell is off to the side rather then center frame"). And I'm not saying they're getting better at writing prompts (which is what they sometimes claim) but just making, for lack of a better word, "tasteful" decisions about what to show and not to show.
What I'm saying is AI users, despite their best efforts, are getting better at "arting"; but the "tool" they claimed made "art" so easy is not actually, it still creates fairly inferior pieces people don't want (most people seem disappointed to learn stuff is AI, even when they otherwise agree "not bad", in my experience; people still want that "bespoke, handcrafted" aspect in their stuff), plus has that whole dubious ethics thing, and, like crypto before it, is coming to light that it uses so much computing power as to be harmful to the environment, oh, also the early adopters spent a lot of time gloating "this is going to straight up kill artists, isn't that great?" which didn't exactly endear them to people. (Also, note, Lowd's a writer, but she sure as shit ain't using generative AI text; my predictions about generative AI having trouble with "sustained" media seems to have been right on the money. You still hear of a dubious attempt by some corporate news organization trying to replace their reporters with AI from time to time, but this never works and is getting even less frequent as time goes by. The only place I've heard it actually being a relatively common problem is student essays, which weren't exactly a form known for producing well crafted writings to begin with.)
I think the reason we're seeing better AI images is not because the programs are getting better, but the users are getting better at ignoring the chimeras and finding the one and only generated image that is halfway decent (and sometimes it's more along the lines of "obvious tell is off to the side rather then center frame"). And I'm not saying they're getting better at writing prompts (which is what they sometimes claim) but just making, for lack of a better word, "tasteful" decisions about what to show and not to show.
What I'm saying is AI users, despite their best efforts, are getting better at "arting"; but the "tool" they claimed made "art" so easy is not actually, it still creates fairly inferior pieces people don't want (most people seem disappointed to learn stuff is AI, even when they otherwise agree "not bad", in my experience; people still want that "bespoke, handcrafted" aspect in their stuff), plus has that whole dubious ethics thing, and, like crypto before it, is coming to light that it uses so much computing power as to be harmful to the environment, oh, also the early adopters spent a lot of time gloating "this is going to straight up kill artists, isn't that great?" which didn't exactly endear them to people. (Also, note, Lowd's a writer, but she sure as shit ain't using generative AI text; my predictions about generative AI having trouble with "sustained" media seems to have been right on the money. You still hear of a dubious attempt by some corporate news organization trying to replace their reporters with AI from time to time, but this never works and is getting even less frequent as time goes by. The only place I've heard it actually being a relatively common problem is student essays, which weren't exactly a form known for producing well crafted writings to begin with.)