Saved in:
| Main Authors: | , , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.17067 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912845141114880 |
|---|---|
| author | Wang, Luozhou Chen, Zhifei Du, Yihua Yan, Dongyu Ge, Wenhang Shen, Guibao Xu, Xinli Wu, Leyi Chen, Man Xu, Tianshuo Ren, Peiran Tao, Xin Wan, Pengfei Chen, Ying-Cong |
| author_facet | Wang, Luozhou Chen, Zhifei Du, Yihua Yan, Dongyu Ge, Wenhang Shen, Guibao Xu, Xinli Wu, Leyi Chen, Man Xu, Tianshuo Ren, Peiran Tao, Xin Wan, Pengfei Chen, Ying-Cong |
| contents | Large-scale video generation models have demonstrated emergent physical coherence, positioning them as potential world models. However, a gap remains between contemporary "stateless" video architectures and classic state-centric world model theories. This work bridges this gap by proposing a novel taxonomy centered on two pillars: State Construction and Dynamics Modeling. We categorize state construction into implicit paradigms (context management) and explicit paradigms (latent compression), while dynamics modeling is analyzed through knowledge integration and architectural reformulation. Furthermore, we advocate for a transition in evaluation from visual fidelity to functional benchmarks, testing physical persistence and causal reasoning. We conclude by identifying two critical frontiers: enhancing persistence via data-driven memory and compressed fidelity, and advancing causality through latent factor decoupling and reasoning-prior integration. By addressing these challenges, the field can evolve from generating visually plausible videos to building robust, general-purpose world simulators. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_17067 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | A Mechanistic View on Video Generation as World Models: State and Dynamics Wang, Luozhou Chen, Zhifei Du, Yihua Yan, Dongyu Ge, Wenhang Shen, Guibao Xu, Xinli Wu, Leyi Chen, Man Xu, Tianshuo Ren, Peiran Tao, Xin Wan, Pengfei Chen, Ying-Cong Computer Vision and Pattern Recognition Artificial Intelligence Large-scale video generation models have demonstrated emergent physical coherence, positioning them as potential world models. However, a gap remains between contemporary "stateless" video architectures and classic state-centric world model theories. This work bridges this gap by proposing a novel taxonomy centered on two pillars: State Construction and Dynamics Modeling. We categorize state construction into implicit paradigms (context management) and explicit paradigms (latent compression), while dynamics modeling is analyzed through knowledge integration and architectural reformulation. Furthermore, we advocate for a transition in evaluation from visual fidelity to functional benchmarks, testing physical persistence and causal reasoning. We conclude by identifying two critical frontiers: enhancing persistence via data-driven memory and compressed fidelity, and advancing causality through latent factor decoupling and reasoning-prior integration. By addressing these challenges, the field can evolve from generating visually plausible videos to building robust, general-purpose world simulators. |
| title | A Mechanistic View on Video Generation as World Models: State and Dynamics |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2601.17067 |