Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.10553 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912930715402240 |
|---|---|
| author | Yuan, Jianhao Zhang, Xiaofeng Friedrich, Felix Beltran-Velez, Nicolas Hall, Melissa Askari-Hemmat, Reyhane Han, Xiaochuang Ballas, Nicolas Drozdzal, Michal Romero-Soriano, Adriana |
| author_facet | Yuan, Jianhao Zhang, Xiaofeng Friedrich, Felix Beltran-Velez, Nicolas Hall, Melissa Askari-Hemmat, Reyhane Han, Xiaochuang Ballas, Nicolas Drozdzal, Michal Romero-Soriano, Adriana |
| contents | State-of-the-art video generative models produce promising visual content yet often violate basic physics principles, limiting their utility. While some attribute this deficiency to insufficient physics understanding from pre-training, we find that the shortfall in physics plausibility also stems from suboptimal inference strategies. We therefore introduce WMReward and treat improving physics plausibility of video generation as an inference-time alignment problem. In particular, we leverage the strong physics prior of a latent world model (here, VJEPA-2) as a reward to search and steer multiple candidate denoising trajectories, enabling scaling test-time compute for better generation performance. Empirically, our approach substantially improves physics plausibility across image-conditioned, multiframe-conditioned, and text-conditioned generation settings, with validation from human preference study. Notably, in the ICCV 2025 Perception Test PhysicsIQ Challenge, we achieve a final score of 62.64%, winning first place and outperforming the previous state of the art by 7.42%. Our work demonstrates the viability of using latent world models to improve physics plausibility of video generation, beyond this specific instantiation or parameterization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_10553 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Inference-time Physics Alignment of Video Generative Models with Latent World Models Yuan, Jianhao Zhang, Xiaofeng Friedrich, Felix Beltran-Velez, Nicolas Hall, Melissa Askari-Hemmat, Reyhane Han, Xiaochuang Ballas, Nicolas Drozdzal, Michal Romero-Soriano, Adriana Computer Vision and Pattern Recognition State-of-the-art video generative models produce promising visual content yet often violate basic physics principles, limiting their utility. While some attribute this deficiency to insufficient physics understanding from pre-training, we find that the shortfall in physics plausibility also stems from suboptimal inference strategies. We therefore introduce WMReward and treat improving physics plausibility of video generation as an inference-time alignment problem. In particular, we leverage the strong physics prior of a latent world model (here, VJEPA-2) as a reward to search and steer multiple candidate denoising trajectories, enabling scaling test-time compute for better generation performance. Empirically, our approach substantially improves physics plausibility across image-conditioned, multiframe-conditioned, and text-conditioned generation settings, with validation from human preference study. Notably, in the ICCV 2025 Perception Test PhysicsIQ Challenge, we achieve a final score of 62.64%, winning first place and outperforming the previous state of the art by 7.42%. Our work demonstrates the viability of using latent world models to improve physics plausibility of video generation, beyond this specific instantiation or parameterization. |
| title | Inference-time Physics Alignment of Video Generative Models with Latent World Models |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2601.10553 |