Wonder is a general-purpose video world model for real-time, camera-controllable exploration. Starting from an image or conditional video, it creates a playable world in which users can move the camera, discover unseen regions, and revisit prior areas over long rollouts. The system combines a dense coordinate-field camera condition, sparse-attention memory for scalable retrieval, and modifications to self-forcing-style distillation. The paper claims diverse, minute-scale generation at 16 FPS while preserving coherent geometry, appearance, and dynamics. It also supports video-conditioned re-shooting of existing dynamic scenes.
No heat snapshots are available in the last 24 hours.