mlsanityData to 14 Oct 2026

Do generative video models understand physical principles?

arXiv 2501.09038 ↗First posted 14 Jan 2025cs.CV cs.AI cs.GRcited by 161

Not on the map. Why?

Cited by 54 in World models, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

AI video generation is undergoing a revolution, with quality and realism advancing rapidly. These advances have led to a passionate scientific debate: Do video models learn "world models" that discover laws of physics -- or, alternatively, are they merely sophisticated pixel predictors that achieve visual realism without understanding the physical principles of reality? We address this question by developing Physics-IQ, a comprehensive benchmark dataset that can only be solved by acquiring a deep understanding of various physical principles, like fluid dynamics, optics, solid mechanics, magnetism and thermodynamics. We find that across a range of current models (Sora, Runway, Pika, Lumiere, Stable Video Diffusion, and VideoPoet), physical understanding is severely limited, and unrelated to visual realism. At the same time, some test cases can already be successfully solved. This indicates that acquiring certain physical principles from observation alone may be possible, but significant challenges remain. While we expect rapid advances ahead, our work demonstrates that visual realism does not imply physical understanding. Our project page is at https://physics-iq.github.io; code at https://github.com/google-deepmind/physics-IQ-benchmark.

Key facts. Cited by 161 papers in the map (data through 14 Oct 2026). Builds on 30 of its 73 references that are in the map. First posted on arXiv 14 Jan 2025.

Builds on

30 of 73 references.

References · 30

  1. 2018The Unreasonable Effectiveness of Deep Features as a Perceptual Metriccited by 11,177
  2. 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,224
  3. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
  4. 2023Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasetscited by 2,581
  5. 2022Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language modelscited by 1,998
  6. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  7. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  8. 2023VBench: Comprehensive Benchmark Suite for Video Generative Modelscited by 1,822
  9. 2018Towards Accurate Generative Models of Video: A New Metric & Challengescited by 1,234
  10. 2025Cosmos World Foundation Model Platform for Physical AIcited by 864
  11. 2023VideoPoet: A Large Language Model for Zero-Shot Video Generationcited by 492
  12. 2019CLEVRER: CoLlision Events for Video REpresentation and Reasoningcited by 484
18 more
  1. 2024Lumiere: A Space-Time Diffusion Model for Video Generationcited by 455
  2. 2022Using cognitive psychology to understand GPT-3cited by 398
  3. 2023A Survey of Hallucination in Large Foundation Modelscited by 368
  4. 2018Generalisation in humans and deep neural networkscited by 237
  5. 2024How Far is Video Generation from World Model: A Physical Law Perspectivecited by 212
  6. 2024VideoScore: Building Automatic Metrics to Simulate Fine-grained Human Feedback for Video Generationcited by 197
  7. 2016Learning to Poke by Poking: Experiential Learning of Intuitive Physicscited by 154
  8. 2024Vision language models are blind: Failing to translate detailed visual features into wordscited by 147
  9. 2016A Compositional Object-Based Approach to Learning Physical Dynamicscited by 119
  10. 2019CoPhy: Counterfactual Learning of Physical Dynamicscited by 81
  11. 2023Physion++: Evaluating Physical Scene Understanding that Requires Online Inference of Different Physical Propertiescited by 54
  12. 2024On the Content Bias in Fr'echet Video Distancecited by 52
  13. 2020CRAFT: A Benchmark for Causal Reasoning About Forces and inTeractionscited by 44
  14. 2023Visual cognition in multimodal large language modelscited by 38
  15. 2023GRASP: A novel benchmark for evaluating language GRounding And Situated Physics understanding in multimodal language modelscited by 28
  16. 2024PhysGame: Uncovering Physical Commonsense Violations in Gameplay Videoscited by 22
  17. 2021Tiered Reasoning for Intuitive Physics: Toward Verifiable Commonsense Language Understandingcited by 19
  18. 2020ESPRIT: Explaining Solutions to Physical Reasoning Taskscited by 16

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 49; 2026: 112.

161 papers.

Sort:

By area:

  1. 2025MAGI-1: Autoregressive Video Generation at Scalecited by 276
  2. 2025Video models are zero-shot learners and reasonerscited by 220
  3. 2025World Simulation with Video Foundation Models for Physical AIcited by 200
  4. 2025FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Drivingcited by 166
  5. 2025WISA: World Simulator Assistant for Physics-Aware Text-to-Video Generationcited by 90
  6. 2025WoW: Towards a World omniscient World model Through Embodied Interactioncited by 67
  7. 2025Back to the Features: DINO as a Foundation for Video World Modelscited by 58
  8. 2025IntPhys 2: Benchmarking Intuitive Physics Understanding In Complex Synthetic Environmentscited by 52
  9. 2025Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systemscited by 48
  10. 2025Reflective Planning: Vision-Language Models for Multi-Stage Long-Horizon Robotic Manipulationcited by 42
  11. 2025RoboScape: Physics-informed Embodied World Modelcited by 40
  12. 2026Inference-time Physics Alignment of Video Generative Models with Latent World Modelscited by 37
38 more
  1. 2025"PhyWorldBench": A Comprehensive Evaluation of Physical Realism in Text-to-Video Modelscited by 36
  2. 2025VaViM and VaVAM: Autonomous Driving through Video Generative Modelingcited by 35
  3. 2025CausalVQA: A Physically Grounded Causal Reasoning Benchmark for Video Modelscited by 35
  4. 2025AI-Generated Video Detection via Perceptual Straighteningcited by 34
  5. 2025PAI-Bench: A Comprehensive Benchmark For Physical AIcited by 34
  6. 2025What about gravity in video generation? Post-Training Newton's Laws with Verifiable Rewardscited by 25
  7. 2026Flow caching for autoregressive video generationcited by 25
  8. 2026Interpreting Physics in Video World Modelscited by 22
  9. 2025Reasoning Physical Video Generation with Diffusion Timestep Tokens via Reinforcement Learningcited by 18
  10. 2025TRAVL: A Recipe for Making Video-Language Models Better Judges of Physics Implausibilitycited by 18
  11. 2025Think Before You Diffuse: Infusing Physical Rules into Video Diffusioncited by 16
  12. 2025Precise Action-to-Video Generation Through Visual Action Promptscited by 15
  13. 2026Physical Simulator In-the-Loop Video Generationcited by 14
  14. 2026Physion-Eval: Evaluating Physical Realism in Generated Video via Human Reasoningcited by 14
  15. 2026Code as Worlds: Agentic Discovery of Executable World Representations for Physical Reasoningcited by 14
  16. 2025Enhancing Physical Plausibility in Video Generation by Reasoning the Implausibilitycited by 13
  17. 2026Wow, wo, val! A Comprehensive Embodied World Model Evaluation Turing Testcited by 13
  18. 2026WorldBench: Benchmarking Physical Understanding of World Models by Isolating Physics Conceptscited by 12
  19. 2026NEWTON: Agentic Planning for Physically Grounded Video Generationcited by 12
  20. 2025ProPhy: Progressive Physical Alignment for Dynamic World Simulationcited by 11
  21. 2026PhysicsMind: Sim and Real Mechanics Benchmarking for Physical Reasoning and Prediction in Foundational VLMs and World Modelscited by 11
  22. 20254DWorldBench: A Comprehensive Evaluation Framework for 3D/4D World Generation Modelscited by 10
  23. 2026RISE-Video: Can Video Generators Decode Implicit World Rules?cited by 10
  24. 2026ACWM-Phys: Investigating Generalized Physical Interaction in Action-Conditioned Video World Modelscited by 10
  25. 2026VOID: Video Object and Interaction Deletioncited by 9
  26. 2026Video Models Can Reason with Verifiable Rewardscited by 9
  27. 2026Physics-IQ Verifiedcited by 9
  28. 2025Learning 3D-Gaussian Simulators from RGB Videoscited by 8
  29. 2025Video-GPT via Next Clip Diffusioncited by 8
  30. 2026MAD: Motion Appearance Decoupling for efficient Driving World Modelscited by 8
  31. 2026World2Act: Latent Action Post-Training from World Model Dynamicscited by 8
  32. 2026Do Joint Audio-Video Generation Models Understand Physics?cited by 8
  33. 2026What-If World: A Causal Benchmark for General World Models in Embodied Scenarioscited by 8
  34. 2025MotionRAG: Motion Retrieval-Augmented Image-to-Video Generationcited by 7
  35. 2025Learning Plug-and-play Memory for Guiding Video Diffusion Modelscited by 7
  36. 2026VisPhyWorld: Probing Physical Reasoning via Code-Driven Video Reconstructioncited by 7
  37. 2026Evolution of Video Generative Foundationscited by 7
  38. 2025MORSE-500: A Programmatically Controllable Video Benchmark to Stress-Test Multimodal Reasoningcited by 6

50 most cited of 161 All 161 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026PhyGround: Benchmarking Physical Reasoning in Generative World Models
  2. 2026PhysInOne: Visual Physics Learning and Reasoning in One Suite
  3. 2025LikePhys: Evaluating Intuitive Physics Understanding in Video Diffusion Models via Likelihood Preference
  4. 2025Exploring the Evolution of Physics Cognition in Video Generation: A Survey
  5. 2026PhysicsLENS: Diagnosing Physical Property Blindness in Video Generation Models
  6. 2025PhysCtrl: Generative Physics for Controllable and Physics-Grounded Video Generation
  7. 2025PhysMaster: Mastering Physical Representation for Video Generation via Reinforcement Learning
  8. 2024Compositional Physical Reasoning of Objects and Events from Videos