T2VPhysBench: A First-Principles Benchmark for Physical Consistency in Text-to-Video Generation
Xuyang Guo, Jiayan Huo, Zhenmei Shi, Zhao Song, Jiahao Zhang, Jiale Zhao
TL;DR
T2VPhysBench introduces a human-evaluated, first-principles benchmark to assess whether text-to-video systems obey fundamental physical laws. It tests 12 laws across Newtonian mechanics, conservation principles, and phenomenological effects using 84 prompts and evaluates 10 models with a rigorous scoring protocol. Results show universal physics violations, especially for conservation laws, and reveal that prompt hints or even counterfactual prompts do not reliably induce correct physics, indicating reliance on pattern memorization rather than true physical reasoning. The work highlights the need for physics-aware video generation and outlines potential directions, including world foundation models and physics-informed learning, to improve real-world physical coherence in generated videos.
Abstract
Text-to-video generative models have made significant strides in recent years, producing high-quality videos that excel in both aesthetic appeal and accurate instruction following, and have become central to digital art creation and user engagement online. Yet, despite these advancements, their ability to respect fundamental physical laws remains largely untested: many outputs still violate basic constraints such as rigid-body collisions, energy conservation, and gravitational dynamics, resulting in unrealistic or even misleading content. Existing physical-evaluation benchmarks typically rely on automatic, pixel-level metrics applied to simplistic, life-scenario prompts, and thus overlook both human judgment and first-principles physics. To fill this gap, we introduce \textbf{T2VPhysBench}, a first-principled benchmark that systematically evaluates whether state-of-the-art text-to-video systems, both open-source and commercial, obey twelve core physical laws including Newtonian mechanics, conservation principles, and phenomenological effects. Our benchmark employs a rigorous human evaluation protocol and includes three targeted studies: (1) an overall compliance assessment showing that all models score below 0.60 on average in each law category; (2) a prompt-hint ablation revealing that even detailed, law-specific hints fail to remedy physics violations; and (3) a counterfactual robustness test demonstrating that models often generate videos that explicitly break physical rules when so instructed. The results expose persistent limitations in current architectures and offer concrete insights for guiding future research toward truly physics-aware video generation.