aiqbench analysis
WorldCrafter advances video world models by introducing an implicit 3D-aware memory mechanism that maintains long-horizon consistency across varying viewpoints without relying on explicit depth correspondences. This approach allows for streaming scene exploration from single inputs while effectively managing token budgets through pose-conditioned readout modules. The method represents a significant step toward robust, interactive virtual environments driven by advanced generative AI techniques.
WorldCrafter: Consistent Video World Model with Implicit 3D-aware Memory
arXiv:2609.24984v1 Announce Type: cross Abstract: Video world models enable interactive exploration of dynamic environments, yet struggle to respect prior observations over long horizons and across viewpoints. We present WorldCrafter, a video world model that learns a camera-queryable implicit 3D-aware memory for this purpose. The key insight is to let the requested viewpoint shape how multi-view evidence is compressed into the video generator's limited token budget. Trained jointly with the video generator, a memory encoder and pose-conditioned readout module integrate historical observations into a fixed set of target view-specific tokens before denoising, without explicit depth-based correspondences. By combining this memory with recent temporal context and few-step distillation, WorldCrafter enables streaming scene exploration from a single input image or text prompt. Experiments across static and dynamic scenes show substantial gains in long-horizon consistency and camera-control accuracy while preserving visual quality during minute-scale exploration.
Abstract:Video world models enable interactive exploration of dynamic environments, yet struggle to respect prior observations over long horizons and across viewpoints. We present WorldCrafter, a video world model that learns a camera-queryable implicit 3D-aware memory for this purpose. The key insight is to let the requested viewpoint shape how multi-view evidence is compressed into the video generator's limited token budget. Trained jointly with the video generator, a memory encoder and pose-conditioned readout module integrate historical observations into a fixed set of target view-specific tokens before denoising, without explicit depth-based correspondences. By combining this memory with recent temporal context and few-step distillation, WorldCrafter enables streaming scene exploration from a single input image or text prompt. Experiments across static and dynamic scenes show substantial gains in long-horizon consistency and camera-control accuracy while preserving visual quality during minute-scale exploration.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Graphics (cs.GR)
arXiv:2609.24984 [cs.CV]
(or arXiv:2609.24984v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.24984 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 238 words
Related coverage
CRePE: Curved Ray Expectation Positional Encoding for Unified-Camera-Controlled Video Generation
arXiv · Sep 9, 2026
Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy — Jason Ma, Dyna Robotics
Latent Space · Oct 13, 2026
Beyond Pixel Histories: World Models with Persistent 3D State
arXiv · Jun 5, 2026
$\omega$-EVA: Envision, Verify, and Act with Latent Interactive World Models
arXiv · Jun 9, 2026