That is true, and ideally this would be a tool for artists (I would love to save time on backgrounds and things for example, using AI to fill in the parts of work I find tedious and time consuming and just fix it up as needed)

Yeah, under a socialist system this wouldn't even be a question, this would be a uniformly wonderful tool that would enable a level and scale of arts production never before imagined provided some selection was put into place as to what could actually get published. Like I've been poking at setting up a rotoscoping pipeline to see if it's at all possible to bridge quick and janky CGI and traditional cell animation with basic hand rotoscoping and an AI cleanup and detailing pass followed by interpolation over multiple frames with ebsynth or something, but I need a new stylus because the battery in mine died and it isn't one where that's replaceable.

unfortunately it also doesn't generate wholly new art, it creates a collage of existing work,

This is inaccurate: it's much, much fuzzier than that, and is more about picking up and recombining concepts and aesthetics - it's weird and repetitive, but tends to be repetitive in the same way artists can be when they work out an approach to a pose and then just keep doing slight variations on it even when that doesn't make sense (like old comics did this a lot) or when they're sticking too close to a reference image. Toss in controlnet and guide it more and it breaks away from that.

but doesn't attribute any of the art of the other artists used to make it. So even if it were a tool used by artists, it would be effectively stealing art from other artists in the process.

I think the clearest refutation of the property angle is to look at two things: who has the power to claim ownership over the training data (hosting sites, major corporations, and social media sites) and whether or not the training data only being properly licensed ahead of time would make a difference in the harm the technology causes. Like who profits if we end up saying "AI trainers must pay royalties to the proper institutions"? Reddit, imgur, meta, deviantart, tumblr, etc, all of whom claim ownership over their users' posts and are already selling that access, because as far as they're concerned it's not the artists being infringed upon but a misuse of their hosting services. Similarly, if Disney or the like came out with an AI trained on its own private library of works and began replacing animators with it and renting it out to selected studios would that make it ethical? Of course not: it is unethical because of who uses it (techbros and corporations) and its consequences (devaluing skilled labor), not because it violates property rights.

And the problem ultimately is any art space that allows AI art to be used is flooded with it. Look at the front page of deviantart for an example. Used to be an actually interesting art website with unique and interesting stuff, now it is just the same generic hundreds of pieces of AI art because it takes 0 effort to actually make. The market gets flooded and actual artists can't be seen by potential supporters because those supporters would have to wade through a mountain of shit to find their work. So it actively becomes detrimental to artists in any space it is allowed, which is why the only people who can get any sort of use out of it have nothing but contempt for art and artists.

Yep, and it's only going to get worse. Some solution to the AI art spam on social media will have to be found, but even worse is what the use of generative AI in professional environments is going to do. Animators are already overworked and underpaid, and that's only going to get worse when these tools get integrated into their workflows and one worker ends up expected to do the work of what now would be an entire team.

That's why I'm focusing on what this is: an extremely powerful and destructive piece of capital that already exists. We can't stop it from existing or stop capitalists from making things worse with it, all we can do is seize upon it and find ways to use it ourselves - that is try to predict how it's going to be put to work professionally and use it to enable and empower smaller independent teams of artists to do with consumer grade hardware what would previously have required a full studio with many millions of dollars worth of invested capital to accomplish.

In practice it's gonna be like a bigger version of what happened with the advent of easily accessible 3d rendering tools: that shit was truly awful and it infested everything, but gradually the low-grade stuff has become mostly filtered out and some professionals have emerged who actually use the medium well. Since there's no putting it back in the bag, the only thing left is to try to exploit it and springboard off it to new heights however possible.