It would be cool to integrate a face tracker to measure features, then you could learn the latent vectors for things like eye position and then you could drag the eyes around.