Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.31336 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916066242854912 |
|---|---|
| author | Yang, Zhenhao Wu, Xiaoshi Lv, Zhengyao Shi, Xiaoyu Wang, Xintao Wan, Pengfei Gai, Kun Wong, Kwan-Yee K. |
| author_facet | Yang, Zhenhao Wu, Xiaoshi Lv, Zhengyao Shi, Xiaoyu Wang, Xintao Wan, Pengfei Gai, Kun Wong, Kwan-Yee K. |
| contents | Recent advances in video generative models have promoted rapid progress in controllable world models. However, maintaining fine-grained spatio-temporal consistency under long-horizon reasoning remains a key challenge. In this work, we move beyond explicit 3D memory and coarse frame-level implicit modeling, and propose a fine-grained, learnable, and scalable memory for consistent world generation. We first identify two fundamental limitations of naïve learnable memory architectures in long-horizon extrapolation, namely computational inefficiency and attention dispersion. Through a systematic analysis of attention dispersion, we propose DecMem, a decoupled memory architecture that employs Sparse Global Memory for efficient fine-grained access to global history and Anchored Local Memory for stable and high-quality extrapolation. Extensive experiments demonstrate that DecMem significantly outperforms current state-of-the-art methods. By ensuring precise and efficient long-term memory and achieving superior extrapolation capabilities, DecMem enables minute-level controllable long video generation with high fidelity and consistency. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_31336 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | DecMem: Towards Minute-Long Consistent World Generation with Decoupled Memory Yang, Zhenhao Wu, Xiaoshi Lv, Zhengyao Shi, Xiaoyu Wang, Xintao Wan, Pengfei Gai, Kun Wong, Kwan-Yee K. Computer Vision and Pattern Recognition Recent advances in video generative models have promoted rapid progress in controllable world models. However, maintaining fine-grained spatio-temporal consistency under long-horizon reasoning remains a key challenge. In this work, we move beyond explicit 3D memory and coarse frame-level implicit modeling, and propose a fine-grained, learnable, and scalable memory for consistent world generation. We first identify two fundamental limitations of naïve learnable memory architectures in long-horizon extrapolation, namely computational inefficiency and attention dispersion. Through a systematic analysis of attention dispersion, we propose DecMem, a decoupled memory architecture that employs Sparse Global Memory for efficient fine-grained access to global history and Anchored Local Memory for stable and high-quality extrapolation. Extensive experiments demonstrate that DecMem significantly outperforms current state-of-the-art methods. By ensuring precise and efficient long-term memory and achieving superior extrapolation capabilities, DecMem enables minute-level controllable long video generation with high fidelity and consistency. |
| title | DecMem: Towards Minute-Long Consistent World Generation with Decoupled Memory |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2605.31336 |