Video World Models · 2026-10-19
Momentum: Persistent Dynamics for Video World Models
Nedko Savov, Stefan Maria Ailuro, Danda Pani Paudel, Yilun Du, Xi Wang, Luc Van Gool
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
Momentum is a two-stage video world-model pipeline for persistent dynamics beyond the target camera view. It represents an evolving scene as synchronized multi-camera videos from graph-connected reference rigs (8 cameras each). WorldUpdate generates the evolving reference-video “world state,” and WorldInteract grounds target camera video generation in that evolving memory so moving content can reappear consistently outside view.
Why it matters
The authors address a failure mode of persistent video world models: dynamic content leaving the field of view can be forgotten or reappear inconsistently. Momentum uses graph-propagated, temporally synchronized multi-view video to build a time-evolving shared state that can ground new camera trajectories, improving dynamic scene consistency and visual fidelity in CARLA-generated street scenes.
Method
- Scene state as persistent spatiotemporal memory: graph of reference-camera rigs (8 cameras each) with synchronized video used as a time-indexed world state.
- WorldUpdate (WorldUpdate/ Multi-Rig DGWM): joint multi-view generation per rig with conditioning from already-generated neighboring rigs via geometric reprojection and graph propagation.
- WorldInteract (Dynamically Grounded WM): camera-controlled video generation conditioned on observation history, camera trajectory, and time-matched reference video views from the evolving world rollout.
Limitation
Generating reference videos introduces an additional inference stage, trading computation for a shared representation of scene dynamics beyond the current view.
Abstract (from arXiv)
Video world models must capture how a scene evolves beyond the current camera view to generate consistent observations when its content reappears. Existing approaches rely on observation history, static point clouds, or image caches constructed from past frames. These representations are often expensive and, more importantly, cannot update dynamic scene content outside the field of view: generated vehicles and pedestrians may be forgotten after leaving the field of view or reappear in physically inaccurate states. We represent an evolving scene through synchronized videos from reference-camera rigs, each comprising eight cameras that jointly cover the surrounding scene. We connect these rigs in a graph and propagate video generation between neighboring rigs to coordinate scene dynamics across locations and time. We introduce Momentum, a two-stage pipeline: WorldUpdate jointly generates reference-camera videos as a shared memory of scene dynamics, and WorldInteract grounds camera-controlled video generation in this evolving memory, allowing moving content to remain represented outside the target view and reappear consistently. To support training and evaluation, we develop a CARLA pipeline that captures synchronized videos from reference camera rigs and a moving target camera, together with an evaluation framework measuring consistency within the generated world state and the target observations.Experiments on generated street scenes show that grounding target videos in scene dynamics generated by Momentum improves dynamic scene consistency by 126.2% in unseen scenes over static image conditioning, while also improving visual fidelity. Our models and source code will be made publicly available.
Related papers
- WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models
- World-T3: Test-Time Training as Persistent Memory for Real-Time Video World Models
- Reading, Writing, Sharing, and Interacting with the World through Video
- StateCraft: Augmenting Video World Models with Abstract-State Reasoning
- LOCI: Spatial Linear Memory for Streaming World Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-16 · Indexed by Pub-AI: 2026-10-19 UTC
Momentum: Persistent Dynamics for Video World Models
Preprint posted: 2026-10-16 · Original record ↗
- arxiv
- 2610.23423
Explicit source versions (1)
- v1 · 2026-10-16
- Preprint available
Source date: 2026-10-16 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Video world models must capture how a scene evolves beyond the current camera view to generate consistent observations when its content reappears. Existing approaches rely on observation history, static point clouds, or image caches constructed from past frames. These representations are often expensive and, more importantly, cannot update dynamic scene content outside the field of view: generated vehicles and pedestrians may be forgotten after leaving the field of view or reappear in physically inaccurate states. We represent an evolving scene through synchronized videos from reference-camera rigs, each comprising eight cameras that jointly cover the surrounding scene. We connect these rigs in a graph and propagate video generation between neighboring rigs to coordinate scene dynamics across locations and time. We introduce Momentum, a two-stage pipeline: WorldUpdate jointly generates reference-camera videos as a shared memory of scene dynamics, and WorldInteract grounds camera-controlled video generation in this evolving memory, allowing moving content to remain represented outside the target view and reappear consistently. To support training and evaluation, we develop a CARLA pipeline that captures synchronized videos from reference camera rigs and a moving target camera, together with an evaluation framework measuring consistency within the generated world state and the target observations.Experiments on generated street scenes show that grounding target videos in scene dynamics generated by Momentum improves dynamic scene consistency by 126.2% in unseen scenes over static image conditioning, while also improving visual fidelity. Our models and source code will be made publicly available.