Ex-Omni is a new open-source AI system that can understand text or speech and then talk back while moving a 3D face in sync with the voice.
Diffusion language models (dLLMs) can write text in any order, but common decoding methods still prefer left-to-right, which wastes their superpower.