Wan (@Alibaba_Wan) “Introducing our latest research: Wan-Streamer — an end-to-end omni-modal underst”

2026.07.10 06:13

Introducing our latest research: Wan-Streamer — an end-to-end omni-modal understanding and generation model designed for real-time duplex interaction. One model. Listens, watches, understands, and responds with synchronized speech and video. Just a single native-streaming Transformer that runs a full duplex video conversation at ~550ms total latency. Because generation is not tied to a fixed avatar rig, the model can embody anything describable in natural language — human characters, pets, anime figures, and more — all in real-time conversation. New in v0.2: higher resolution (640×368, 25 FPS), ~200ms model-side latency.

显示更多