Editing AI output is actually one of the easier things to do, since you can literally just mask off part of an image and tell it to try again. The most you'd ever have to do is draw a crude sketch of what you want to go there (probably a limb that was missing). The barrier of entry for running stable diffusion locally, at least financially, is probably somewhere around $300 for a used graphics card made within the last 4 years or so, and maybe lower than that. You can get a lot more done on more expensive hardware and with more specific technical ability, sure, but... that's not completely new. The tools for using it are not as user friendly as they could be, but they're at a point right now where probably anyone who can do digital art could start doing AI art well within a few days. And knowing how to draw on your own is extremely helpful and speeds up the process massively, so people who are already artists have a huge advantage.

Honestly I think too much of the AI art discourse is rooted in misunderstanding. I don't think we can shove AI generation back into Pandora's box, and I haven't seen any serious proposal contradicting that, so I'm mostly concerned with mitigating impacts and having people get the most benefit possible out of it. I like to inform people about what it can do, because I sincerely believe it can be an overall beneficial tool, but half the time I get screamed at by people who are just in fight-or-flight mode over it, and who also tend to believe that a straight text prompt can generate anything that isn't just shitpost quality when it simply doesn't work like that. I can understand why, but it doesn't make it any less tiring.