The latest example can combine a single image and an audio file to create a video portrait of someone singing or talking. It can even tweak their emotions.