Reply to thread

Message: <blockquote data-quote="Ruin Explorer" data-source="post: 9313912" data-attributes="member: 18">That's sad to hear and it certainly helps to explain why attitudes to art and artists in the US are more "insane" than most of the world. You don't typically see European posters attempting to imply that they've never had an opportunity to create art, that cruel artists are basically stealing away all the artistic talent and unfairly refusing to work without pay and so on, but that's exactly the sentiments you see from some of AI art's biggest boosters on, say, Twitter.The issue here is really simple - the vast majority of AI art that we are asked to see as "good looking" or told "doesn't look like AI art" very obviously doesn't actually meet those criteria. As I've pointed out, it tends to have two characteristics:1) It doesn't actually look "good", it looks "detailed" - these are different things. But most AI art people are claiming "looks great" tends to merely be detailed, and to broadly match their prompt. Often it looks absolutely horrid - but there's backslapping going on between AI art fans assuring each other it looks great. We can't pretend that isn't happening - especially given it's been happening since the days when Cthulhu Mythos-esque hands were basically the norm. We've been told pieces "look great" and are "indistinguishable from real art" by people posting 22-fingered monstrosities with boobs the size of a basketball, and you expect people to just magically agree it's good now the monstrosity only has 10 fingers most of the time?Again, as I've already pointed out, there is AI art that does look good - but you have to generate dozens or more of images to get there - and most people stop a long way before that, instead getting something that looks kind of "bleh" or even "ugh" but matches their prompt, is brightly coloured, and is detailed, and labelling it "good". And in a sense it is, to them. Not actually good - but "good enough".2) It's not typically as hard to distinguish as people make out, especially not when you have large images. There will absolutely be stuff where you simply have no idea, sure. But there are still specific tells which aren't just mistakes - overdetail, and use of multiple different shading techniques are the classic two. Ironically if AI art wants greater acceptance, they probably need to recalibrate it to produce less detailed images, not more. I don't think that's actually possible, however, with the current technical approach.I think a lot of people, especially some small-press RPG creators, would really like AI art to be harder to spot than it is, simply so they could get away with using it without people looking into it. At this point some of the "false claims" are so obviously dumb they look people intentionally falsely claiming art is AI, so they can act out a little morality play as they pretend to be surprised to find out it wasn't and then handwring and start talking about witch trials. But I suspect it's nothing so cunning - it's just those people are dimwits.I've discussed this and you've just ignored what I've said in favour of bland generalizations, which let me be clear - are very unhelpful to your argument.The reality is, these tools are simply an evolutionary dead end. That's what a lot of people don't seem to understand. AI art will do something pretty amazing at some point, but not with the current scrape-and-compute approach, or whatever you want to call it. The current approach isn't one of creating art, it's one of creating images computationally. It can be refined - errors can be worked out - at the cost of making it more computationally expensive in most cases. But fundamentally the very way it works limits it.I don't know for sure what the non-dead-end product is, but it's probably something that instead of trying to crunch images instantly into existence, actually builds them up - likely with input from a human during the process. I think that's likely to produce much more persuasive results, and much less likely to a dead end that's already up against the wall of what it can do.</blockquote>

Verification