As an aside, it would be cool if music could also be an input to these kinds of generative models, such that the generated image somehow matches the feeling or mood of the music.