Microsoft dropped VASA-1. This AI can make single image sing and talk from audio reference expressively. Here are some examples.