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

TL;DR

This paper introduces a decoupled memory control framework for long-horizon video generation that improves spatial consistency and efficiency by separating memory from the generative process.

Contribution

It proposes a lightweight, independent memory module with cross-attention and camera-aware gating to enhance spatial consistency and reduce training costs.

Findings

Achieves state-of-the-art spatial consistency in generated videos.

Reduces training costs compared to entangled memory models.

Enhances the ability to explore novel scene regions.

Abstract

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to mediate the interaction between memory and generation modules, enabling memory conditioning only when meaningful historical references exist.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.