



I got something more like a displacement effect working, but I don't like the feel of the kind of smooth circle erase versus the crunchiness of everything else. Think I'm going to cut it.
A nano-banana canvas experiment. Generating based on a source image and the overlapping prompts, caching the results. Like adding semantic layers to an image.
Caching by combined image and prompt works well, but doesn't fully solve the order issue of:

Implemented rough cache in the GIF viewer, since GIFs mainly work by applying patches on top of the starting image. I run through the whole thing caching evenly spaced frames at the start, then use them as shortcut base when seeking to frames further in.

Fun to see how he thinks about this stuff - lots of notes about what would be 'cheating' and trying to sort out what the fundamental, interesting challenges are from the incidental challengs.