OpenMOSS

Open-source 32B MoE foundation model that generates synchronized video and audio in one inference step with precise dual-tower lip-sync.