Do generative video models understand physical principles?

TL;DR

This paper introduces Physics-IQ, a benchmark dataset to evaluate whether current generative video models understand physical principles, revealing that visual realism does not equate to physical understanding and highlighting existing limitations.

Contribution

The paper presents Physics-IQ, a new benchmark dataset designed to assess physical understanding in video models, and evaluates several models showing limited physical comprehension despite realistic visuals.

Findings

Models show limited understanding of physical principles.

Physical understanding is largely unrelated to visual realism.

Some physical tasks can be successfully solved by current models.

Abstract

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.