Mode Seeking meets Mean Seeking for Fast Long Video Generation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866910035446071296 |
|---|---|
| author | Cai, Shengqu Nie, Weili Liu, Chao Berner, Julius Zhang, Lvmin Ma, Nanye Chen, Hansheng Agrawala, Maneesh Guibas, Leonidas Wetzstein, Gordon Vahdat, Arash |
| author_facet | Cai, Shengqu Nie, Weili Liu, Chao Berner, Julius Zhang, Lvmin Ma, Nanye Chen, Hansheng Agrawala, Maneesh Guibas, Leonidas Wetzstein, Gordon Vahdat, Arash |
| contents | Scaling video generation from seconds to minutes faces a critical bottleneck: while short-video data is abundant and high-fidelity, coherent long-form data is scarce and limited to narrow domains. To address this, we propose a training paradigm where Mode Seeking meets Mean Seeking, decoupling local fidelity from long-term coherence based on a unified representation via a Decoupled Diffusion Transformer. Our approach utilizes a global Flow Matching head trained via supervised learning on long videos to capture narrative structure, while simultaneously employing a local Distribution Matching head that aligns sliding windows to a frozen short-video teacher via a mode-seeking reverse-KL divergence. This strategy enables the synthesis of minute-scale videos that learns long-range coherence and motions from limited long videos via supervised flow matching, while inheriting local realism by aligning every sliding-window segment of the student to a frozen short-video teacher, resulting in a few-step fast long video generator. Evaluations show that our method effectively closes the fidelity-horizon gap by jointly improving local sharpness, motion and long-range consistency. Project website: https://primecai.github.io/mmm/. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_24289 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Mode Seeking meets Mean Seeking for Fast Long Video Generation Cai, Shengqu Nie, Weili Liu, Chao Berner, Julius Zhang, Lvmin Ma, Nanye Chen, Hansheng Agrawala, Maneesh Guibas, Leonidas Wetzstein, Gordon Vahdat, Arash Computer Vision and Pattern Recognition Machine Learning Scaling video generation from seconds to minutes faces a critical bottleneck: while short-video data is abundant and high-fidelity, coherent long-form data is scarce and limited to narrow domains. To address this, we propose a training paradigm where Mode Seeking meets Mean Seeking, decoupling local fidelity from long-term coherence based on a unified representation via a Decoupled Diffusion Transformer. Our approach utilizes a global Flow Matching head trained via supervised learning on long videos to capture narrative structure, while simultaneously employing a local Distribution Matching head that aligns sliding windows to a frozen short-video teacher via a mode-seeking reverse-KL divergence. This strategy enables the synthesis of minute-scale videos that learns long-range coherence and motions from limited long videos via supervised flow matching, while inheriting local realism by aligning every sliding-window segment of the student to a frozen short-video teacher, resulting in a few-step fast long video generator. Evaluations show that our method effectively closes the fidelity-horizon gap by jointly improving local sharpness, motion and long-range consistency. Project website: https://primecai.github.io/mmm/. |
| title | Mode Seeking meets Mean Seeking for Fast Long Video Generation |
| topic | Computer Vision and Pattern Recognition Machine Learning |
| url | https://arxiv.org/abs/2602.24289 |