I was just fantasizing about how to do stuff like this. In my fantasy there were two DALL-E-like models. One for generating characters and one for generating

I was just fantasizing about how to do stuff like this. In my fantasy there were two DALL-E-like models. One for generating characters and one for generating scenes and a system for adding a generated character to a scene. The user describes a character to the character model and gets candidate illustrations. The user can then save a character by giving it a name and use the name to generate variations of the same character. I'm imagining that either you just have a lot of sliders to vary the vector that the character is drawn from or maybe it's possible to combine ideas (e.g. Character X + running). Then the user just illustrates the scenes they want, the characters, combines them together, and gets a useful image.