LikePhys: Evaluating Intuitive Physics Understanding in Video Diffusion Models via Likelihood Preference
This work addresses the problem of accurately assessing physics understanding in AI models for researchers and developers aiming to build physically plausible world simulators, representing an incremental improvement in evaluation methods.
The paper tackles the challenge of evaluating intuitive physics understanding in video diffusion models by introducing LikePhys, a training-free method that uses a denoising objective as a likelihood surrogate to distinguish physically valid from impossible videos, showing strong alignment with human preference and outperforming state-of-the-art evaluators on a benchmark of twelve scenarios across four physics domains.
Intuitive physics understanding in video diffusion models plays an essential role in building general-purpose physically plausible world simulators, yet accurately evaluating such capacity remains a challenging task due to the difficulty in disentangling physics correctness from visual appearance in generation. To the end, we introduce LikePhys, a training-free method that evaluates intuitive physics in video diffusion models by distinguishing physically valid and impossible videos using the denoising objective as an ELBO-based likelihood surrogate on a curated dataset of valid-invalid pairs. By testing on our constructed benchmark of twelve scenarios spanning over four physics domains, we show that our evaluation metric, Plausibility Preference Error (PPE), demonstrates strong alignment with human preference, outperforming state-of-the-art evaluator baselines. We then systematically benchmark intuitive physics understanding in current video diffusion models. Our study further analyses how model design and inference settings affect intuitive physics understanding and highlights domain-specific capacity variations across physical laws. Empirical results show that, despite current models struggling with complex and chaotic dynamics, there is a clear trend of improvement in physics understanding as model capacity and inference settings scale.