Advancing Narrative Long Video Generation via Training-Free Identity-Aware Memory

2026-05-18

AI Summary

This work addresses the challenges of identity drift, character duplication, and attribute loss in long-form video generation with multiple prompt switches. To mitigate these issues, the authors propose IAMFlow, a novel framework that introduces, for the first time, an explicit, training-free mechanism for identity modeling and tracking. The approach leverages a large language model to extract visual-attribute-rich entities and assign them global identifiers, employs a vision-language model for asynchronous intra-frame attribute verification, and ensures cross-prompt consistency through adaptive prompt transition. By moving beyond conventional implicit similarity matching, IAMFlow achieves a 2.56-point lead over the strongest baseline on the newly introduced NarraStream-Bench benchmark and demonstrates a 1.39× speedup in inference for 60-second multi-prompt scenarios.

Abstract

Autoregressive video generation has improved rapidly in visual fidelity and interactivity, but it still suffers from long-term inconsistency and memory degradation. Most existing solutions either compress historical frames using predefined strategies or retrieve keyframes based on coarse implicit attention signals, both of which fail to handle evolving prompts with shifting entity references, leading to identity drift, character duplication, and attribute loss. To address this, we propose IAMFlow, a training-free identity-aware memory framework that explicitly models and tracks persistent entity identities, enabling consistent generation across prompt transitions. Specifically, an LLM extracts entities with visual attributes from each prompt and assigns unique global IDs for identity-aware memory, while a VLM asynchronously verifies and refines attributes from rendered frames, enabling explicit entity tracking in place of implicit similarity-based matching. To keep the proposed framework computationally practical, we design a systematic inference acceleration pipeline, including asynchronous visual verification, adaptive prompt transition, and model quantization, which achieves faster generation than existing baselines. Furthermore, we introduce NarraStream-Bench, a benchmark for narrative streaming video generation that features 324 multi-prompt scripts spanning six dimensions and a three-dimensional evaluation protocol that integrates both traditional metrics and multimodal large language model-based assessments. Extensive experiments show that IAMFlow, despite being training-free, achieves the best overall performance on NarraStream-Bench, outperforming the strongest baseline by 2.56 points, while achieving a 1.39$\times$ speedup over the most efficient baseline in the 60-second multi-prompt setting.

Problem

• long-term consistency
• identity drift
• autoregressive video generation
• memory degradation
• entity tracking

Innovation

• identity-aware memory
• training-free video generation
• entity tracking
• autoregressive video generation
• narrative video benchmark