Clips2Story creates short videos from shuffled clips with keyword input

Clips2Story: Training-Free Video Storyboarding and Editing Using Multimodal Retrieval-Embedded Generation

Computer Vision and Pattern RecognitionArtificial IntelligenceMachine LearningMultimedia

Summary

When video editors have many small clips out of order, it can be hard to make a good story from them. This paper presents Clips2Story, a tool that does not need training and uses a large language model (LLM) to pick clips related to a keyword and put them in a logical storyboard with narration. The tool then combines the clips and matching images to produce a short, understandable video. The authors show that Clips2Story works well through tests and human judgments, making it easier to create videos without needing the original sequence of footage.

What this means in practice

  • For video editors: Create short thematic videos from unsorted clip collections based on user keywords without extensive training or manual sequencing.
  • For marketing content creators: Automatically assemble keyword-focused video stories from diverse footage sources for promotional or campaign videos.

Authors

Junnuo Wang, Weihan Xu, Hao-Wen Dong

Abstract

Video editors often work with large collections of fragmented footage and must select and organize relevant clips into a coherent story. However, many existing video editing and summarization methods assume a temporally ordered source video and aim to preserve its original narrative and temporal structure. In this paper, we introduce Clips2Story, a training-free, LLM-based pipeline that enables users to create keyword-specified short videos from a pool of video clips. Given a pool of clips and a user-specified keyword, we first extract text-based multimodal descriptors for each clip and retrieve the clips most relevant to the keyword. We then provide the textual descriptors of these candidate clips to an LLM, which selects and arranges them into an editable storyboard with connective narrations. Finally, we assemble the selected clips and retrieve accompanying visuals for the narrations to compose the output video. To evaluate the proposed pipeline, we construct a synthetic evaluation dataset from 100 long-form videos across five genres by segmenting each video into short clips using a scene-change detection model and shuffling the resulting clips to remove their temporal order. Through objective evaluation and qualitative examples, we demonstrate that Clips2Story can effectively retrieve keyword-relevant clips and compose coherent storyboards based on user-specified keywords. Human evaluation further shows that Clips2Story outperforms a strong baseline in terms of narrative coherence, keyword alignment, and overall realism.