# FlowNar: Scalable Streaming Narration for Long-Form Videos

Source: [FlowNar: Scalable Streaming Narration for Long-Form Videos](<https://www.sota2.com/research/paper/6a1e63c5dbd7150506f853b6>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-02T05:00:51.557Z

Authors: Zeyun Zhong, Manuel Martin, Chengzhi Wu, David Schneider, Frederik Diederichs, Juergen Gall, Juergen Beyerer

Published: 2026-05-30T08:51:28.000Z

## Abstract

Recent Large Multimodal Models (LMMs), primarily designed for offline settings, are ill-suited for the dynamic requirements of streaming video. While recent online adaptations improve real-time processing, they still face critical scalability challenges, with resource demands typically growing at least linearly with video duration. To overcome this bottleneck, we propose FlowNar, a novel framework for scalable streaming video narration. The core of FlowNar is a dynamic context management strategy for historical visual context removal, combined with our CLAM (Cross Linear Attentive Memory) module for streaming visual history retention, ensuring bounded visual memory usage and computational complexity, crucial for efficient streaming. We also introduce a realistic self-conditioned evaluation protocol and complementary evaluation metrics to assess streaming narration models under deployment-like conditions. Experiments on the Ego4D, EgoExo4D, and EpicKitchens100 datasets demonstrate that FlowNar substantially improves narration quality over strong baselines while being highly efficient, supporting processing of 10$\\times$ longer videos and achieving 3$\\times$ higher throughput (FPS). The code is available at https://github.com/zeyun-zhong/FlowNar.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Streaming Narration on Ego4D](<https://www.sota2.com/research/sota/streaming-narration-on-ego4d.md>): Streaming Narration; Ego4D; Perplexity (PPL): 1.995.

- [Streaming Narration on Ego4D v1 (test)](<https://www.sota2.com/research/sota/streaming-narration-on-ego4d-v1-test.md>): Streaming Narration; Ego4D v1 (test); MACs: 16.5.

- [Streaming Narration on EgoExo4D v1 (test)](<https://www.sota2.com/research/sota/streaming-narration-on-egoexo4d-v1-test.md>): Streaming Narration; EgoExo4D v1 (test); MACs: 20.39.

- [Streaming Narration on EK100 v1 (test)](<https://www.sota2.com/research/sota/streaming-narration-on-ek100-v1-test.md>): Streaming Narration; EK100 v1 (test); MACs: 24.09.

- [Video Summarization on Ego4D v1 (test)](<https://www.sota2.com/research/sota/video-summarization-on-ego4d-v1-test.md>): Video Summarization; Ego4D v1 (test); CIDEr: 11.2.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2606.00620>)

[PDF](<https://arxiv.org/pdf/2606.00620.pdf>)