SlotMem: Character-Addressable Internal Memory for Narrative Long Video Generation

2026-07-17

AI Summary

This work addresses the challenge of maintaining character identity consistency across scenes and over extended temporal spans in narrative long-form video generation. To this end, the authors propose a character-addressable internal memory framework that employs semantic character probes to locate relevant visual tokens, compresses DiT tokens into compact character-slot memories, and continuously updates and precisely injects these representations during generation. The designed character-slot mechanism enables fine-grained, character-specific memory storage and retrieval, effectively disentangling identity-critical features from incidental visual variations while supporting continual updates under limited memory budgets. Experimental results demonstrate that the proposed method significantly improves long-range character consistency across multiple long-video generation benchmarks, without compromising generation quality relative to existing approaches.

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.

Problem

• character consistency
• long video generation
• narrative video
• identity preservation
• temporal coherence

Innovation

• character-addressable memory
• slot memory
• long video generation
• identity consistency
• cross-attention localization