preprint · arXiv (2026)
Wan-Streamer v0.2: Higher Resolution, Same Latency
Wan Team, Alibaba Group
In plain English
Raises the video output of a real-time audio-visual conversation model from 192x336 to 640x368 at 25 FPS while keeping about 200 ms model-side latency (about 550 ms including an assumed 350 ms network budget). Enables wider, scene-aware mid-shot agents.
real-time avatars
latency
digital humans
conversational ai