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

Jinzhuo Liu1 , Jiangning Zhang1 , Wencan Jiang1 , Yabiao Wang1,2 , Dingkang Liang3 , Zhucun Xue1 , Ran Yi4 , Yong Liu1
1Zhejiang University , 2Tencent Youtu Lab , 3Huazhong University of Science and Technology , 4Shanghai Jiao Tong University
Paper Code Model

Highlights

(1) We introduce IAMFlow , a training-free identity-aware memory framework that explicitly organizes historical information around persistent entities and attributes, enabling reliable identity preservation across evolving prompt transitions.

(2) We design a systematic inference acceleration pipeline , to make the framework computationally practical, combining asynchronous visual verification, adaptive prompt transition, and model quantization to preserve long-term consistency without sacrificing generation speed.

(3) We introduce NarraStream-Bench , a modern benchmark suite for assessing long-term consistency in narrative streaming video generation. Extensive experiments and ablation studies demonstrate that IAMFlow achieves superior performance across various metrics while enabling more efficient inference.

IAMFlow teaser

Demo Videos

IAMFlow
MemFlow
LongLive
Deep Forcing
Rolling Forcing
Self Forcing

Method: IAMFlow

Framework of IAMFlow

Framework of IAMFlow

Benchmark: NarraStream-Bench

Overview of NarraStream-Bench

Overview of NarraStream-Bench

Benchmark Comparison

Comparison of related long-video generation benchmarks. The custom VBench-Long protocol only supports metrics for visual quality.

Benchmark VQ TC IC Prompt Type Aggregation Strategy Year
VBench-Long ✓ × × Single Slow-Fast Avg. 2024
LV-Bench ✓ ✓ × Single VDE 2025
NarrLV × ✓ ✓ Single TNA-based QA 2025
NarraStream-Bench ✓ ✓ ✓ Multi Narrative-Aware 2026

Experiments

Qualitative Visualization

Qualitative results of IAMFlow under NarraStream-Bench

Quantitative Comparison

Quantitative comparison for the multi-prompt 60-second setting on NarraStream-Bench. Self Forcing, Rolling Forcing, and Deep Forcing are adapted for the experiment. Best results are in bold, and second-best results are underlined.

ID Method Visual Quality ↑ Temporal Consistency ↑ Instruction Compliance ↑ Overall ↑
SC BC TF MS VTSS BS CAC CLC EG DT VLM
1 Self Forcing 80.85 82.05 86.29 56.17 30.39 60.25 50.57 39.54 55.63 51.98 50.96 56.71
2 Rolling Forcing 88.16 85.93 80.55 41.41 52.59 61.19 68.83 62.18 58.45 56.37 52.03 63.14
3 Deep Forcing 92.27 90.74 88.54 50.61 73.99 48.10 80.09 77.28 60.71 62.87 59.35 69.57
4 LongLive 95.36 93.32 90.27 57.76 77.29 49.19 83.06 89.97 62.40 63.33 62.22 73.17
5 MemFlow 95.37 93.21 90.23 57.37 77.67 47.06 83.92 89.71 61.93 64.14 60.85 72.88
6 IAMFlow 95.88 93.43 90.36 59.49 77.56 65.03 86.07 91.25 62.03 64.56 62.61 75.73

BibTeX

@misc{liu2026advancingnarrativelongvideo,
  title={Advancing Narrative Long Video Generation via Training-Free Identity-Aware Memory},
  author={Jinzhuo Liu and Jiangning Zhang and Wencan Jiang and Yabiao Wang and Dingkang Liang and Zhucun Xue and Ran Yi and Yong Liu},
  year={2026},
  eprint={2605.18733},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2605.18733},
}