Wonder: Video World Model Done Better

2026-07-28

AI Summary

This work proposes an interactive, real-time method for modeling video worlds with long-horizon camera controllability, enabling users to freely explore unseen regions and revisit previously observed scenes within the generated environment. Through a system-level co-design, the approach introduces a dense coordinate field–based camera conditioning mechanism, a sparse-attention memory module, and an enhanced self-distillation training strategy, collectively enhancing the model’s responsiveness to control signals and its long-term temporal consistency. The method supports multimodal generation tasks—including image-to-video synthesis and video reenactment—while maintaining geometric, appearance, and motion coherence over durations of up to several minutes and achieving real-time performance at 16 frames per second.

Abstract

We present Wonder, a general-purpose video world model for real-time, camera-controllable world exploration. Given an image or a conditional video, Wonder constructs a playable world where users can navigate interactively by moving the camera, discovering unseen regions, and revisiting previously observed areas in real time and over a long-term horizon. Achieving this capability requires a system-level co-design of control method, memory mechanism, and training strategy. We introduce a novel camera conditioning with a dense coordinate field whose renderings provide spatially aligned motion and orientation cues, allowing the model to interpret camera motion directly as visual evidence. To support fast and precise memory retrieval over a growing generation context, we propose an efficient sparse attention-based memory mechanism, enabling the model to selectively attend to a small set of relevant context tokens at inference time, regardless of actual context length. We further develop several techniques to rectify the self-forcing-style distillation pipeline, improving the student model's ability to respect control signals, as well as maintaining diverse generation modes and long-term memory from the teacher. Together, these components enable Wonder to synthesize diverse, minute-scale videos at 16 FPS while preserving coherent geometry, appearance, and dynamics across long rollouts. Beyond image-to-video generation, Wonder naturally supports video-conditioned generation, allowing existing dynamic scenes to be re-shot in real time.

Problem

• video world model
• camera-controllable generation
• long-term coherence
• interactive exploration
• real-time video synthesis

Innovation

• video world model
• camera conditioning
• sparse attention memory
• real-time controllable generation
• self-forcing distillation