VideoPhy-2: A Challenging Action-Centric Physical Commonsense Evaluation in Video Generation
read the original abstract
Large-scale video generative models, capable of creating realistic videos of diverse visual concepts, are strong candidates for general-purpose physical world simulators. However, their adherence to physical commonsense across real-world actions remains unclear (e.g., playing tennis, backflip). Existing benchmarks suffer from limitations such as limited size, lack of human evaluation, sim-to-real gaps, and absence of fine-grained physical rule analysis. To address this, we introduce VideoPhy-2, an action-centric dataset for evaluating physical commonsense in generated videos. We curate 200 diverse actions and detailed prompts for video synthesis from modern generative models. We perform human evaluation that assesses semantic adherence, physical commonsense, and grounding of physical rules in the generated videos. Our findings reveal major shortcomings, with even the best model achieving only 22% joint performance (i.e., high semantic and physical commonsense adherence) on the hard subset of VideoPhy-2. We find that the models particularly struggle with conservation laws like mass and momentum. Finally, we also train VideoPhy-AutoEval, an automatic evaluator for fast, reliable assessment on our dataset. Overall, VideoPhy-2 serves as a rigorous benchmark, exposing critical gaps in video generative models and guiding future research in physically-grounded video generation. The data and code is available at https://videophy2.github.io/.
This paper has not been read by Pith yet.
Forward citations
Cited by 39 Pith papers
-
PhysInOne: Visual Physics Learning and Reasoning in One Suite
PhysInOne is a new dataset of 2 million videos across 153,810 dynamic 3D scenes covering 71 physical phenomena, shown to improve AI performance on physics-aware video generation, prediction, property estimation, and m...
-
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
MemoBench is a new diagnostic benchmark with 360 synthetic and real clips plus VQA evaluation that tests memory consistency in video models under the disappear-and-reappear paradigm in dynamically changing environments.
-
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
MemoBench curates 360 ground-truth clips and an evaluation suite to diagnose memory consistency failures in video models when objects change state while out of view.
-
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
MemoBench is a new diagnostic benchmark with automated and VQA metrics that evaluates memory consistency in video models under disappear-and-reappear in dynamic environments.
-
Each Judge Its Own Yardstick: Discovering Per-VLM Taxonomies for Physical Video Evaluation
JudgeFit produces per-VLM physical video evaluation taxonomies that improve held-out accuracy by a mean 32% relative to a single global schema across 16 models from eight families.
-
YoCausal: How Far is Video Generation from World Model? A Causality Perspective
YoCausal benchmark shows video diffusion models detect the arrow of time but lack genuine causal understanding relative to humans.
-
What-If World: A Causal Benchmark for General World Models in Embodied Scenarios
What-If World is a new paired-prompt benchmark showing that nine state-of-the-art video generation models achieve at most 52% on causal intervention tests and cluster near 28% for open-source systems.
-
WBench: A Comprehensive Multi-turn Benchmark for Interactive Video World Model Evaluation
WBench is a benchmark with 289 test cases and 1,058 turns for evaluating interactive world models using 22 automated metrics validated against human judgments.
-
CRONOS: Benchmarking Counterfactual Physical Consistency in Video Models
CRONOS benchmark shows recent open-source video generators fail to preserve physical consistency under controlled changes to viewpoint, scene, object category, and appearance.
-
CaC: Advancing Video Reward Models via Hierarchical Spatiotemporal Concentrating
CaC presents a new spatiotemporal concentrating reward model for video anomalies, built on a novel large-scale dataset and three-stage training with RL and IoU rewards, claiming 25.7% accuracy gains and 11.7% anomaly ...
-
CaC: Advancing Video Reward Models via Hierarchical Spatiotemporal Concentrating
CaC is a hierarchical spatiotemporal concentrating reward model for video anomalies that reports 25.7% accuracy gains on fine-grained benchmarks and 11.7% anomaly reduction in generated videos via a new dataset and GR...
-
PhyGround: Benchmarking Physical Reasoning in Generative World Models
PhyGround is a new benchmark with curated prompts, a 13-law taxonomy, large-scale human annotations, and an open physics-specialized VLM judge for evaluating physical reasoning in generative video models.
-
Do Joint Audio-Video Generation Models Understand Physics?
AV-Phys Bench shows that current joint audio-video models lack robust physical commonsense, with major drops on transitions and deliberate anti-physics prompts.
-
Do Joint Audio-Video Generation Models Understand Physics?
Current joint audio-video generation models lack robust physical commonsense, especially during transitions and when prompted for impossible behaviors.
-
BRITE: A Benchmark for Reliable and Interpretable T2V Evaluation on Implausible Scenarios
BRITE benchmark reveals that leading T2V models handle static object composition well but degrade sharply on object-action binding and audio-visual synchronization for implausible prompts.
-
MoRight: Motion Control Done Right
MoRight disentangles object and camera motion via canonical-view specification and temporal cross-view attention, while decomposing motion into active user-driven and passive consequence components to learn and apply ...
-
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
MemoBench curates 360 clips and an evaluation suite to test video models on recovering updated object states after disappear-and-reappear in changing environments.
-
Current World Models Lack a Persistent State Core
Current world models fail to evolve internal state when unobserved and instead resume scenes at the last observed state, as diagnosed by the new WRBench benchmark across 23 models and 9600 videos.
-
Proprio: Latent Self-Scoring and Inference-Time Refinement for Physically Plausible Video Generation
Proprio uses flow residuals from latent perturbations in frozen video generators as a self-scoring signal for physical plausibility, yielding reported gains of 16.5% on Physics-IQ and 20.6% on VideoPhy2-hard.
-
LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation
LaMo adds self-supervised latent motion priors via a motion drift loss during training and motion prior guidance during sampling to boost physical fidelity in video diffusion models like CogVideoX.
-
NEWTON: Agentic Planning for Physically Grounded Video Generation
NEWTON improves physical accuracy in video generation by deploying a trainable planner that coordinates physics-aware tools and a verifier, raising joint accuracy on VideoPhy-2 without altering the base generators.
-
SARA: Semantically Adaptive Relational Alignment for Video Diffusion Models
SARA improves text alignment and motion quality in video diffusion models by routing token-relation distillation supervision to semantically salient pairs using a Stage-1 aligner trained with SAM masks and InfoNCE.
-
SARA: Semantically Adaptive Relational Alignment for Video Diffusion Models
SARA introduces semantic saliency to guide relational alignment in video diffusion models, improving text following and motion quality over prior alignment methods.
-
How Far Are Video Models from True Multimodal Reasoning?
Current video models succeed on basic understanding but achieve under 25% success on logically grounded generation and near 0% on interactive generation, exposing gaps in multimodal reasoning.
-
Self-Refining Video Sampling
Self-refining video sampling treats a pre-trained generator as a denoising autoencoder for iterative inference-time refinement guided by self-consistency uncertainty to improve motion coherence and physics alignment.
-
ProPhy: Progressive Physical Alignment for Dynamic World Simulation
ProPhy adds explicit physics-aware conditioning via semantic and refinement experts plus VLM knowledge transfer to produce more physically coherent dynamic videos than prior methods.
-
PhyDetEx: Detecting and Explaining the Physical Plausibility of T2V Models
A new dataset and fine-tuned VLM detector/explainer called PhyDetEx shows that current T2V models still struggle to generate videos that obey physical laws, with open-source models performing worse.
-
Enhancing Physical Plausibility in Video Generation by Reasoning the Implausibility
A training-free framework uses physics-violating counterfactual prompts and Synchronized Decoupled Guidance to suppress implausible motions in diffusion-based video generation while preserving photorealism.
-
PhysRAG: Enhancing Physics-Awareness in Video Generation via Retrieval-Augmented Generation
PhysRAG curates 7K videos from WISA-80K, builds a physical video database, and injects knowledge via learnable queries into a diffusion model to reach SOTA visual quality and physical compliance on PhyGenBench and VBench.
-
Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment
PILA aligns frozen flow-matching video models to a physics attribute bank via MoE experts and operational residuals, reporting SOTA physical plausibility on VBench-2.0, VideoPhy-2 and PhyGenBench while preserving visu...
-
MPMWorlds: Material-Point-Method Simulations for Inferring and Extrapolating Physical Dynamics
Assembles MPM simulation dataset and compares code generation versus video diffusion for inferring physical parameters and extrapolating dynamics from videos.
-
OptiWorld: Optimal Control for Video World Generation under Physical Constraints
OptiWorld inserts a classical optimal-control layer that extracts a world state, plans an optimal trajectory on a geometric manifold under physical constraints, and renders the video conditioned on that trajectory.
-
Tempered Self-Similarity Alignment for Physically Plausible Video Generation
Tempered Self-similarity Alignment transfers relational structure from foundation-model STSS into video generators via probabilistic correspondence alignment, yielding reported gains in physical plausibility on VideoP...
-
PhyWorld: Physics-Faithful World Model for Video Generation
PhyWorld improves temporal consistency and physical plausibility in video world models via flow matching fine-tuning followed by DPO on physics preference pairs, with reported gains on VBench and a custom physical-fai...
-
DriveCtrl: Conditioned Sim-to-Real Driving Video Generation
DriveCtrl is a depth-conditioned controllable framework that generates realistic driving videos from simulation while preserving annotations and scene dynamics.
-
Agentic Physical AI toward a Domain-Specific Foundation Model for Nuclear Reactor Control
A compact language model trained on scaled synthetic nuclear reactor control data exhibits variance collapse and emergent concentration on a single actuation strategy driven by physical execution success.
-
MASS: Motion-Aware Spatial-Temporal Grounding for Physics Reasoning and Comprehension in Vision-Language Models
MASS adds spatiotemporal motion signals and 3D grounding to VLMs and releases MASS-Bench, yielding physics-reasoning performance within 2% of Gemini-2.5-Flash after reinforcement fine-tuning.
-
World Simulation with Video Foundation Models for Physical AI
Cosmos-Predict2.5 unifies text-to-world, image-to-world, and video-to-world generation in one model trained on 200M clips with RL post-training, delivering improved quality and control for physical AI.
-
Evolution of Video Generative Foundations
This survey traces video generation technology from GANs to diffusion models and then to autoregressive and multimodal approaches while analyzing principles, strengths, and future trends.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.