"PhyWorldBench": A Comprehensive Evaluation of Physical Realism in Text-to-Video Models

arXiv 2025 42.7 benchmark

TLDR

PhyWorldBench evaluates physics realism in text-to-video models using 1050 prompts across 12 models, with human and multimodal LLM evaluation.

Reasoning

The paper presents a comprehensive benchmark and evaluation methodology, covering diverse physical phenomena and including an anti-physics category. Its strengths are the broad model coverage and mixed human/MLLM evaluation, though it does not introduce a new generative model and limitations are not detailed in the abstract.

Read-first score

Read-first score 42.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 26.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=analysis,benchmark,evaluation

Topical relevance 42%

37.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 564.

Keyword Scores

video world model

world dynamics prediction

world model

world simulator

generative world model

interactive world model

model-based reinforcement learning world model

Resources

Paper