Co-Edit: Human-AI Collaborative Video Editing With RAG-Based B-Roll Narrative

Video creation has become increasingly popular, but creators report that selecting interesting clips, arranging clips, and inserting suitable B-roll is challenging. In this paper, we investigate how to use human-AI co-creation to make editing easier through suggestions and automation. We propose Co-Edit, a human-AI co-creative system designed to support efficient generation, modifications, and B-roll insertion for video creation. Co-Edit allows users to edit videos through natural language instructions or direct UI manipulation. In addition, we presented a retrieval-augmented generation (RAG)-based B-roll insertion method to enhance content authenticity, which can suggest B-roll content and positions for the edited video. To evaluate the system, we conducted a within-subjects user study with 12 participants. The study demonstrates significant reductions in cognitive load while maintaining creative control, with participants reporting enhanced efficiency in producing platform-optimized content.