I've run the scribbles through an image model to get keywords, captions and styles. This what the search bar is hooked up to.

It's pretty good for searching for things like "still life" or "green".