That makes much more sense, pretty typical usage of ML as a blunt instrument.

The interesting thing to do, I think, would be to have data set of general images, and use a movie description to pull from those images.