DreamID-Omni is one model that can create, edit, and animate human-centered videos with matching voices, all in sync.
This paper teaches talking avatars not just to speak, but to look around their scene and handle nearby objects exactly as a text instruction says.