3) AI still has a long way to go before it can competently and fully do 1) or 2)
I love what it's doing so far, do not get me wrong, but I'm seeing that there are still hiccups in the output, a lack of consistency across multiple images (unless the character is very, VERY bland), and then there's still the specter of Model Collapse, which crops up in private LLMs like a cancer.
Give it 3 years or so, and most of this should be ironed out. 5 years from now, you can download and keep one churning along on a fairly high-performance desktop.