SlotMem: Character-Addressable Internal Memory for Narrative Long Video Generation
Computer Vision and Pattern Recognition
Summary
The authors focus on improving how videos keep the same characters looking consistent over long scenes and time gaps. They point out that previous methods either use memory that doesn't specifically track characters well or mix up character identity with other visual details. Their new method, SlotMem, creates a special character-specific memory that updates carefully as the video is generated, making sure each character’s look stays stable. Tests show SlotMem keeps character consistency better than earlier methods without losing video quality.
Authors
Yilai Liu, Xin Zhang, Shiyuan Zhang, Hongyang Du
Abstract
Maintaining recurring character identities across scene transitions and long temporal gaps is a central challenge in narrative long video generation. Methods targeting global consistency often retrieve memory using cues that are not aligned with character identity preservation, while recent character-centric variants still rely on coarse frame-level kv memory that entangles identity with incidental visual factors and lacks a continuous update mechanism under limited memory capacity. To address these limitations, we propose \textbf{SlotMem}, a character-addressable internal memory framework for multi-character narrative long video generation. Specifically, SlotMem uses a Character-Semantic Probe to localize character-relevant visual tokens from cross-attention responses, and a Memory Encoder to compress DiT tokens into compact role-wise slot memory. As generation proceeds, a Memory Writer conservatively updates each character's memory with new observations, while Character-Wise Cross-Attention retrieves the role memory and injects it only into localized tokens of the same character. Experiments on multiple narrative long video generation benchmarks show that SlotMem improves long-range character consistency over existing baselines, while maintaining comparable video quality. Our code is available at https://github.com/YilaiLiu-HKU/SlotMem.