Kepler4D generates controllable future videos from single camera views

Kepler4D: Controllable Future Video Generation via 4D Scene State Evolution

Computer Vision and Pattern Recognition

Summary

Predicting how objects move in videos is tricky, especially when using just one camera. The authors created Kepler4D, which builds a 3D model of a scene from a single video and tracks object movements over time. This model allows users to guide how objects will move next and then generate future videos that look realistic and consistent with the scene. It separates the big motion decisions from how objects appear, making the future video controllable and believable.

What this means in practice

  • For video game developers: Create dynamic in-game scenes by predicting and controlling object movements from single camera captures.$Commercial implications: Enables game studios to generate realistic future animations from limited footage, enhancing game environment dynamics.
  • For film post-production teams: Generate controllable future actions of scene objects for video editing and visual effects from existing footage.

Authors

Feiran Wang, Bin Duan, Junyi Wu, Gaowen Liu, Yan Yan

Abstract

Video world models aim to preserve scene structure and predict how dynamic objects evolve beyond visual observations. We present Kepler4D, a framework for future video generation through explicit 4D scene state evolution. Given a monocular video, Kepler4D constructs a shared 3D representation of background geometry, object motion histories, coarse spatial supports, and semantic context. Chain-of-Motion summarizes observed motion and uses a vision-language model to select structured speed and heading decisions and decide whether to bound object-center height from below. A deterministic rollout converts these decisions into future object trajectories for inspection and editing before synthesis. We render the evolving proxies into geometric controls for a pretrained video generator, separating coarse object motion from the synthesis of appearance and articulation. Experiments on real-world videos demonstrate that Kepler4D enables controllable object motion and plausible future rollout while preserving scene consistency.