Video World Models · 2026-10-14
A Plausible Future Is Not a Causal Future: Can Video Models Generate How the World Unfolds?
Jialiang Li, Zhengning Xu, Zongyang Ma, Shaobo Cui
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors study whether image-to-video models can generate “causal cascades”: a full chain of causally dependent events. They introduce CausalCascade, a simulator-grounded benchmark with 50 Blender scenes and 2,832 counterfactual variants. Evaluations on nine I2V models show better spatial localization than correct spatiotemporal evolution, improved event recovery with richer prompts but limited physical realization, and weak selective responsiveness to physical interventions.
Why it matters
The paper frames causal cascade generation as a process-level world-model test: not just producing plausible events, but sustaining causally coherent evolution, selectively propagating intervention effects, and maintaining scene consistency. It provides a simulator-grounded benchmark and diagnostic evaluation across motion, event recovery, and counterfactual responsiveness, highlighting gaps that limit current I2V models as faithful causal world models.
Method
- Introduce CausalCascade: 50 event-rich Blender scenes with structured event graphs and graph-grounded diagnostic VQA, plus three future-event specification levels (state-only, trigger-aware, full event-chain).
- Construct 2,832 counterfactual variants via delete/move/scale (RGB-visible) and mass (RGB-invisible) interventions, and resimulate each variant to obtain factual–counterfactual reference rollouts.
- Evaluate nine image-to-video models using motion-level metrics (Spatial IoU, Strict ST-IoU, Weighted Spatial IoU) and counterfactual criteria (Effect and Preserve IoU variants), plus VQA over event occurrence, temporal ordering, causal pathway, and outcome/termination.
Limitation
The authors report that audited failures cluster at foundational stages, where models fail to establish a persistent physical world with stable entities and events that can support sustained causal propagation.
Abstract (from arXiv)
Large-scale video generation models are increasingly viewed as a promising path toward world models, yet their ability to capture causally coherent physical dynamics remains poorly understood. Existing evaluations largely emphasize visual quality, prompt adherence, or broad physical plausibility, leaving open whether a model can faithfully realize a complete causal cascade, in which an initiating event propagates through causally dependent interactions to a terminal outcome. We introduce CausalCascade, a simulator-grounded benchmark centered on 50 event-rich Blender scenes with structured event graphs, graph-grounded diagnostic VQA, three levels of future-event specification, and 2,832 controlled counterfactual variants, and use it to evaluate nine image-to-video models with complementary motion-, event-, and counterfactual-level diagnostics. Across the evaluated models, generated motion is more reliably localized spatially than realized with correct spatiotemporal evolution, while richer future-event specification improves event recovery but leaves physical realization limited. Counterfactual evaluation further reveals weak selective responsiveness to physical interventions, while most audited failures are foundational, arising at scene consistency or event occurrence. Together, these results position causal cascade generation as a process-level test of world modeling and highlight the need to jointly recover the intended future, faithfully realize its physical evolution, selectively respond to physical interventions, and maintain a coherent physical world over time. Code and dataset are available at https://github.com/SJTU-DeepDeltaLab/CausalCascade .
Related papers
- Foresight at the Event Boundary: Evaluating Physical Prediction in Video World Models
- World Models Need a Physical Registry: Learning Internal Predictive States for Physically Plausible World Models
- Reading, Writing, Sharing, and Interacting with the World through Video
- Momentum: Persistent Dynamics for Video World Models
- Time-JEPA: Grounding Video World Models with Physical Time Series
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-11 · Indexed by Pub-AI: 2026-10-14 UTC
A Plausible Future Is Not a Causal Future: Can Video Models Generate How the World Unfolds?
Preprint posted: 2026-10-11 · Original record ↗
- arxiv
- 2610.15562
Explicit source versions (1)
- v1 · 2026-10-11
- Preprint available
Source date: 2026-10-11 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Large-scale video generation models are increasingly viewed as a promising path toward world models, yet their ability to capture causally coherent physical dynamics remains poorly understood. Existing evaluations largely emphasize visual quality, prompt adherence, or broad physical plausibility, leaving open whether a model can faithfully realize a complete causal cascade, in which an initiating event propagates through causally dependent interactions to a terminal outcome. We introduce CausalCascade, a simulator-grounded benchmark centered on 50 event-rich Blender scenes with structured event graphs, graph-grounded diagnostic VQA, three levels of future-event specification, and 2,832 controlled counterfactual variants, and use it to evaluate nine image-to-video models with complementary motion-, event-, and counterfactual-level diagnostics. Across the evaluated models, generated motion is more reliably localized spatially than realized with correct spatiotemporal evolution, while richer future-event specification improves event recovery but leaves physical realization limited. Counterfactual evaluation further reveals weak selective responsiveness to physical interventions, while most audited failures are foundational, arising at scene consistency or event occurrence. Together, these results position causal cascade generation as a process-level test of world modeling and highlight the need to jointly recover the intended future, faithfully realize its physical evolution, selectively respond to physical interventions, and maintain a coherent physical world over time. Code and dataset are available at https://github.com/SJTU-DeepDeltaLab/CausalCascade .