Memorize When Needed: Decoupled Memory Control for Spatially Consistent Long-Horizon Video Generation

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Guo, Yanjun, Zhang, Zhengqiang, Wang, Pengfei, Liang, Xinyue, Ma, Zhiyuan, Zhang, Lei
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866917426469273600
author Guo, Yanjun
Zhang, Zhengqiang
Wang, Pengfei
Liang, Xinyue
Ma, Zhiyuan
Zhang, Lei
author_facet Guo, Yanjun
Zhang, Zhengqiang
Wang, Pengfei
Liang, Xinyue
Ma, Zhiyuan
Zhang, Lei
contents Spatially consistent long-horizon video generation aims to maintain temporal and spatial consistency along predefined camera trajectories. Existing methods mostly entangle memory modeling with video generation, leading to inconsistent content during scene revisits and diminished generative capacity when exploring novel regions, even trained on extensive annotated data. To address these limitations, we propose a decoupled framework that separates memory conditioning from generation. Our approach significantly reduces training costs while simultaneously enhancing spatial consistency and preserving the generative capacity for novel scene exploration. Specifically, we employ a lightweight, independent memory branch to learn precise spatial consistency from historical observation. We first introduce a hybrid memory representation to capture complementary temporal and spatial cues from generated frames, then leverage a per-frame cross-attention mechanism to ensure each frame is conditioned exclusively on the most spatially relevant historical information, which is injected into the generative model to ensure spatial consistency. When generating new scenes, a camera-aware gating mechanism is proposed to mediate the interaction between memory and generation modules, enabling memory conditioning only when meaningful historical references exist. Compared with the existing method, our method is highly data-efficient, yet the experiments demonstrate that our approach achieves state-of-the-art performance in terms of both visual quality and spatial consistency.
format Preprint
id arxiv_https___arxiv_org_abs_2604_18215
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Memorize When Needed: Decoupled Memory Control for Spatially Consistent Long-Horizon Video Generation
Guo, Yanjun
Zhang, Zhengqiang
Wang, Pengfei
Liang, Xinyue
Ma, Zhiyuan
Zhang, Lei
Computer Vision and Pattern Recognition
Spatially consistent long-horizon video generation aims to maintain temporal and spatial consistency along predefined camera trajectories. Existing methods mostly entangle memory modeling with video generation, leading to inconsistent content during scene revisits and diminished generative capacity when exploring novel regions, even trained on extensive annotated data. To address these limitations, we propose a decoupled framework that separates memory conditioning from generation. Our approach significantly reduces training costs while simultaneously enhancing spatial consistency and preserving the generative capacity for novel scene exploration. Specifically, we employ a lightweight, independent memory branch to learn precise spatial consistency from historical observation. We first introduce a hybrid memory representation to capture complementary temporal and spatial cues from generated frames, then leverage a per-frame cross-attention mechanism to ensure each frame is conditioned exclusively on the most spatially relevant historical information, which is injected into the generative model to ensure spatial consistency. When generating new scenes, a camera-aware gating mechanism is proposed to mediate the interaction between memory and generation modules, enabling memory conditioning only when meaningful historical references exist. Compared with the existing method, our method is highly data-efficient, yet the experiments demonstrate that our approach achieves state-of-the-art performance in terms of both visual quality and spatial consistency.
title Memorize When Needed: Decoupled Memory Control for Spatially Consistent Long-Horizon Video Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.18215