Yeah there’s something called “Ai dementia” that’s referring to Ai trained on Ai generated images. It can get better in one aspect, but often gets worse in others. The thing about the image creation though in a world with Ai generated images is that there will always be a big industry of people compositing and making “meta-images” with Ai content, then also the critical factor of quality is merely human judgement.
I think it will be a slight problem with poor datasets, but ultimately an ocean of human selected and composited images will still make the Ai improve. Then also I think there will be a big market for artists who can *fix* the problems (cough* hands cough*) of Ai generated images/animation.
In other words, I think it will only be a problem if people stop caring about what is good and what sucks. Otherwise, I think the larger selection will just help the Ai distinguish good quality from bad.
(Not to mention methodology will probably get an order of magnitude boost in the next few years too. If there’s a downside, we will probably never see it)