Video = World + Event Stream

2026-07-16

AI Summary

This work proposes a general-purpose, real-time multimodal pretraining framework tailored for large-scale real-world videos to support diverse downstream interactive applications. The core innovation lies in decomposing video into a stable contextual “world” and a dynamic “event stream,” which informs a unified real-time prediction task. By integrating multimodal fusion, joint visual-language-action modeling, and a low-latency streaming inference architecture, the system achieves real-time processing at 640×368 resolution and 25 FPS, with an end-to-end interaction latency of approximately 550 ms—including a 350 ms network budget. The approach demonstrates strong transferability to real-time scenarios such as full-duplex audiovisual interaction, significantly enhancing the efficiency of mapping user multimodal inputs to language and action responses.

Abstract

We present Wan-Streamer v0.3, which reframes our native-streaming interaction model under a single organizing view: a video is a world plus an event stream. The world is the persistent context in which a video unfolds, including the environment, scene, subjects, ambient acoustic conditions, voice characteristics, and other relatively stable conditions. The event stream is everything that changes over time within that world, including scene or environmental changes, subject behavior, speech, and other sounds. This yields a general-purpose pretraining task over large amounts of real video: given a world and incoming input, predict how the world moves, changes, and responds in real time. The resulting competence can be specialized to a broad family of real-time downstream tasks. We instantiate it on real-time full-duplex audio-visual interaction, where the event stream is the agent's speech together with free-form behavior. Functionally, the model's multimodal understanding process is vision-language-action-like: it maps multimodal user input to language-form speech and behavior actions. Wan-Streamer v0.3 preserves the v0.2 operating point: 640x368 video at 25 FPS, a 160 ms streaming unit, approximately 200 ms model-side response latency, and approximately 550 ms total interaction latency under a 350 ms bidirectional network budget.

Problem

• video understanding
• event stream
• real-time interaction
• multimodal learning
• world modeling

Innovation

• World-Event Stream Framework
• Real-time Multimodal Interaction
• Streaming Video Pretraining
• Full-duplex Audio-Visual Agent
• Low-latency Video Modeling