Video World Models · 2026-10-15
WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models
Nan Zhou, Jianxiong Gao, Qiuyu Wang, Wen Wang, Jiahao Wang, Yufeng Yuan, Shuaiting Li, Hao Ouyang, Yujun Shen, Weiming Lu, Jun Xiao, Yueting Zhuang, Yongliang Shen
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
WorldMemBench is a benchmark for memory in video world models, covering 260 test cases. It evaluates static memory via human identity/appearance, object appearance/geometry, and environment content/geometry, and dynamic memory via event progression and state persistence under controlled camera departure-and-revisit trajectories. The authors report that strong static retention does not imply reliable dynamic evolution, and that repeated revisits reveal cumulative memory degradation.
Why it matters
The authors argue that visual quality and local temporal coherence cannot establish whether generated worlds can preserve past observations and evolve coherently over time. WorldMemBench provides a unified, systematic evaluation separating static content retention from dynamic state evolution, using fixed-reference multi-revisit protocols and event-gated dynamic checklists, plus video quality and camera-following diagnostics for interpretation.
Method
- Benchmark design: 260 cases (200 static-memory, 60 dynamic-memory) across 23 content categories; static and dynamic memory are evaluated as separate dimensions.
- Controlled departure-and-revisit: translation loops and rotation trajectories with fixed initial references; static revisits are compared against the same initial reference across successive returns.
- Dynamic scoring: each case specifies initial state, action, and expected next state; a VLM gate determines whether the event starts, then eight fixed questions score event progression, persistence/reset, final state, and causal/physical consistency.
Limitation
The current benchmark focuses on controlled translation and rotation trajectories over rollouts of several hundred frames. Longer horizons, free-form or compound traversal patterns, and embodied settings in which camera and object actions are jointly selected remain outside its present scope. Dynamic Memory additionally relies on frozen case-specific checklists evaluated by a VLM, and the Causal/Physics dimension remains the most challenging sub-dimension.
Abstract (from arXiv)
As video models enable longer and more realistic interactive experiences, maintaining a world consistent with its history becomes essential. This requires both preserving previously observed content and ensuring coherent world evolution over time. World memory is a critical capability to evaluate alongside visual fidelity and video duration. We introduce WorldMemBench, a benchmark for systematically evaluating world memory in video world models. Across 260 test cases, it assesses static memory through the retention of human identity and appearance, object appearance and geometry, and environmental content and geometry, and dynamic memory through event progression and state persistence. Using controlled camera departures and multi-revisit trajectories, the benchmark evaluates static content against fixed initial references and dynamic events against expected state transitions. It combines appearance and geometry metrics with event-specific assessments, and reports video quality and camera-following accuracy as complementary diagnostics. Evaluation of nine representative models shows that strong static retention does not imply reliable dynamic state evolution, while repeated revisits reveal memory degradation beyond the first return. Human studies support the alignment between our automatic evaluations and human judgments. WorldMemBench provides a systematic testbed for diagnosing memory failures and guiding the development of persistent, temporally coherent video worlds.
Related papers
- World-T3: Test-Time Training as Persistent Memory for Real-Time Video World Models
- Momentum: Persistent Dynamics for Video World Models
- Reading, Writing, Sharing, and Interacting with the World through Video
- StateCraft: Augmenting Video World Models with Abstract-State Reasoning
- UniState: Unifying Memory and Intervention for Controllable World Simulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-13 · Indexed by Pub-AI: 2026-10-15 UTC
WorldMemBench: A Comprehensive Benchmark for Memory of Video World Models
Preprint posted: 2026-10-13 · Original record ↗
- arxiv
- 2610.19063
Explicit source versions (1)
- v1 · 2026-10-13
- Preprint available
Source date: 2026-10-13 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
As video models enable longer and more realistic interactive experiences, maintaining a world consistent with its history becomes essential. This requires both preserving previously observed content and ensuring coherent world evolution over time. World memory is a critical capability to evaluate alongside visual fidelity and video duration. We introduce WorldMemBench, a benchmark for systematically evaluating world memory in video world models. Across 260 test cases, it assesses static memory through the retention of human identity and appearance, object appearance and geometry, and environmental content and geometry, and dynamic memory through event progression and state persistence. Using controlled camera departures and multi-revisit trajectories, the benchmark evaluates static content against fixed initial references and dynamic events against expected state transitions. It combines appearance and geometry metrics with event-specific assessments, and reports video quality and camera-following accuracy as complementary diagnostics. Evaluation of nine representative models shows that strong static retention does not imply reliable dynamic state evolution, while repeated revisits reveal memory degradation beyond the first return. Human studies support the alignment between our automatic evaluations and human judgments. WorldMemBench provides a systematic testbed for diagnosing memory failures and guiding the development of persistent, temporally coherent video worlds.