I think it might be useful to have an AI art program that, instead of generating the whole image at once, it generates in stages.

Artists usually do a line draft first, to figure out the basic geometry of the drawing. Then they do more detailed line work, coloring, shading, etc.

Essentially a ground-up approach, whereas current AI image gen focuses on starting with noise and iteratively adding details until you get something that mostly doesn't look like shit. Then you edit how the prompt is worded so that the AI has a better idea of what to add or avoid.

What I'm saying is that AI tools should be built to aid the actual art process, rather than being a words game. It's currently designed so that people with no experience in art can make art, but I haven't heard of any tool designed to aid in each stage of the process.