here are some pics from google. anyone who raised a kid or was a kid knows what I Spy books are. :)
I wanted to have the prompt: “A page from an I Spy book about ____” work since it would make totally epic pages and could just switch in different topics and wow. But Perchance has no idea what I Spy books are. Does anyone else’s ai know what I Spy books are?
‘Do a whole bunch of shit at once’ seems unreliable. But you can image-to-image from parts of a whole render, and get specific pieces to come out better, and then fit back into the big picture.
The two elements of useful generation seem to be filtering and editing. Say you’re doing a comic page. You want a particular expression in one panel? Yeah, you’ll have to try several times, maybe doodle over a near miss and refine. You want a particular layout? Well then do panels individually and lay them out; talking the model into a specific pattern is harder and worse.