WorldPlay: Towards Long-Term Geometric Consistency for Real-Time Interactive World Modeling

This paper presents WorldPlay, a streaming video diffusion model that enables real-time, interactive world modeling with long-term geometric consistency, resolving the trade-off between speed and memory that limits current methods. WorldPlay draws power from three key ingredients. 1) We use a Dual Action Representation to enable robust action control in response to the user's keyboard and mouse inputs. 2) To enforce long-term consistency, our Reconstituted Context Memory dynamically rebuilds context from past frames and uses temporal reframing to keep geometrically important but long-past frames accessible, effectively alleviating memory attenuation. 3) We also propose Context Forcing, a novel distillation method designed for memory-aware model. Aligning memory context between the teacher and student preserves the student's capacity to use long-range information, enabling real-time speeds while preventing error drift. Taken together, WorldPlay generates long-horizon streaming 720p video at 24 FPS with superior consistency, comparing favorably with existing techniques and showing strong generalization across diverse scenes. Project page and online demo can be found: https://3d-models.hunyuan.tencent.com/world/ and https://3d.hunyuan.tencent.com/sceneTo3D.

Imagine a virtual world where every observation is generated in real-time by a generative model. You can move freely with your keyboard and mouse, returning to a doorway or landmark you passed and finding it exactly where you left it. Today's generators face a trade-off: those fast enough to feel interactive tend to forget the past and drift, while those that remember well are too slow to play. WorldPlay closes that gap. Three ideas make this possible: a Dual Action Representation turns raw keyboard and mouse signals into robust control actions; a Reconstituted Context Memory acts like a notebook that rebuilds context from past frames so the model doesn't forget where it has been; and Context Forcing, a new distillation technique, accelerates the model to real-time speed without erasing its long-range memory. Together, Worldplay enables long-horizon, navigable virtual worlds — pointing toward interactive simulators, robot training environments, and immersive content tools that anyone can steer in real time.