Especially where image generation is concerned, the infancy part can’t be understated. It’s growing so, so fast. A year ago, people would be dismissing AI art as “you can always tell”, it largely couldn’t do hands, and text was right out. But current cutting edge models can semi-reliably generate imperceptible works, needing only some fairly trivial manual curation to pick the best output. There’s also some models that are now able to do basic text. Just comparing a couple of years worth of progress side by side makes it very clear that it’s advancing rapidly and there’s no signs yet that it’s plateaued.

The big barrier to image generation, though, is profit. The images that it creates are useful, but current understanding is that they can’t be copyrighted and there’s ongoing legal challenges that make it very murky. I don’t think these companies can stay in business from regular people who’ll pay for some tokens to generate art. They need to be usable by commercial companies, and the legal issues will scare many of those away, at least for now.