Auto-rig of facial expressions. I'm ispired by how much potential it has, especially in context of long animations or even cartoons.