Been experimenting with allowing GPT to control the canvas more by prompting it to execute basic canvas functions. Still a very early work in progress, but check out the video in the tweet below and the example starter flow if you want to try it yourself (features are live)!
In this video example, I simply ask a super-prompted chat to:
"help me learn about Art Deco"
- it spawns several image creations it feels are relevant
- opens up 3 chat windows (with itself) to get more info about some sub-topics
- summarizes the 3 chat topics
- and then summarizes the entire process into a single paragraph
All from my simple "I want to learn more" message. All I did is zoom/pan the canvas and move around the results.