There has been an explosion of works in recent years - some good and some lousy. I have a bunch of books on mental imaging and have not been satisfied with them, so I've evolved a special architecture for it I believe is novel in the field. It uses a mental whiteboard for data and modeling and a mind's eye that applies perception viewpoints to extract meaning from it. On the output side I reverse that and the mind's eye applies styles to output to the mental whiteboard. There's much more complication to it than that, but anyway it works very well for generative AI. I live where all the autonomous vehicle r&d is done and I see better approaches than their perception frameworks. Heh, I'm just down the street from Fremont Tesla and used to be near pony.ai and saw their test vehicles all the time on slower streets. This is an exciting time to be here.