So far, it looks like any image created by AI is a one off. And AI is "trained" using existing images. Side note: you can really mess up an AI by including eff'ed up scenarios (in this case, images) into the mix.
So maybe AI can create a cover illustration. But let's say we want to see a Superman story drawn by Leonardo da Vinci (or Charles Schulz for that matter). How would that work ? You load up every image drawn or painted by da Vinci. Then you load up images (like those model sheets) of Superman, his villains and every supporting character. You program human anatomy (e.g. elbows only bend one way). Let's say you go by the "Marvel method". There's a very loose outline of a story. How does the AI pace the story? What about panel layout, # of panels per page, splash pages, 1/2 splashes... establishing shot, medium shot, close up shots... leaving room for lettering balloons and captions. I guess it could be done. But it may not be pretty. Then you go with the more traditional method (full -script)... then what ??