Wan-Streamer v0.2: Higher-Resolution Audio-Visual Interaction Maintains Low Latency
By
[Submitted on 5 Jul 2026]
Summary
Wan-Streamer v0.2 is a latency-preserving upgrade to an end-to-end audio-visual interaction model. It increases the interactive output stream resolution from 192x336 to 640x368 while maintaining approximately 200 ms model-side signal-to-signal latency at 25 FPS. The higher resolution enables scene-grounded mid-shot agents with legible posture, gaze, hands, and scene layout during real-time conversation. The architecture uses a single-GPU "thinker" for low-latency streaming perception and a multi-GPU "performer" group (using Ulysses-style context parallelism) for high-resolution latent video generation, keeping total remote interaction latency at approximately 550 ms including network budget.
Source
bskyWan-Streamer v0.2: Higher-Resolution Audio-Visual Interaction Maintains Low Latencyarxiv.orgKey quotes
· 3 pulledv0.2 keeps the v0.1 modeling formulation, but raises the interactive output stream from 192x336 to 640x368 while preserving approximately 200 ms model-side signal-to-signal latency at 25 FPS.
The higher-resolution stream supports scene-grounded mid-shot agents whose posture, gaze, hands, nearby objects, and local scene layout remain legible during real-time conversation.
This concentrates additional hardware on visual generation while preserving the compact thinker-performer boundary, keeping total remote interaction latency at approximately 550 ms when a 350 ms bidirectional network budget is included.
You might also wanna read
Wan 2.2: First Open-Source MoE Model for AI Video Generation
Wan 2.2 is a major open-source update to the Wan video models, introducing a Mixture-of-Experts (MoE) architecture for enhanced performance
Product Hunt·11mo ago
Higgsfield Launches WAN 2.5 AI Video Generation Platform with Advanced Creative Controls
Higgsfield has launched WAN 2.5, a next-generation AI video generation model that offers creators full creative control with features includ
Product Hunt·1mo ago
Helios: A 14B Parameter Real-Time Video Generation Model for Minute-Scale Content
Helios is a 14B parameter video generation model that achieves real-time performance at 19.5 FPS on a single NVIDIA H100 GPU while supportin
alphaxiv.org·4mo ago
StreamingVLM: Real-Time Vision-Language Model for Infinite Video Stream Processing
StreamingVLM is a new vision-language model designed for real-time understanding of infinite video streams, addressing the computational cha
arxiv.org·8mo ago
Developer Creates PyroWave, a Low-Latency Video Codec for Game Streaming
The article discusses the development of PyroWave, a custom-designed, ultra-low-latency video codec for game streaming. It emphasizes the im
themaister.net·11mo ago
Comments
Sign in to join the conversation.
No comments yet. Be the first.