Do Joint Audio-Video Generation Models Understand Physics?

TL;DR

This paper introduces AV-Phys Bench, a benchmark for evaluating physical commonsense in joint audio-video generation models, revealing current models' limitations in understanding real-world physics and scene dynamics.

Contribution

The paper presents AV-Phys Bench for assessing physical understanding in audio-video models and introduces AV-Phys Agent for improved evaluation aligned with human judgments.

Findings

Seedance 2.0 performs best among tested models.

All models struggle with scene transitions and anti-physics prompts.

Models lack robustness in physical understanding and cross-modal consistency.

Abstract

Joint audio-video generation models are rapidly approaching professional production quality, raising a central question: do they understand audio-visual physics, or merely generate plausible sounds and frames that violate real-world consistency? We introduce AV-Phys Bench, a benchmark for evaluating physical commonsense in joint audio-video generation. AV-Phys Bench tests models across three scene categories: Steady State, Event Transition, and Environment Transition. It covers physics-grounded subcategories drawn from real-world scenes, plus Anti-AV-Physics prompts that deliberately request physically inconsistent audio-video behavior. Each generation is evaluated along five dimensions: visual semantic adherence, audio semantic adherence, visual physical commonsense, audio physical commonsense, and cross-modal physical commonsense. Across three proprietary and four open-source models, we find that Seedance 2.0 performs best overall, but all models remain far from robust physical understanding. Performance drops sharply on event-driven and environment-driven transitions, and even strong proprietary systems collapse on Anti-AV-Physics prompts. We further introduce AV-Phys Agent, a ReAct-style evaluator that combines a multimodal language model with deterministic acoustic measurement tools, producing rankings that closely align with human ratings. Our results identify cross-modal physical consistency and transition-driven scene dynamics as key open challenges for joint audio-video generation.