ReVi: Recursive Recap Video Generation
| Source: arXiv AI
Tags: ReVi, video generation, multi-agent systems, recap video, narrative summarization, text-to-video
ReVi is a multi-agent system that turns a full novel into a short recap video within a fixed time budget, planning story, narration and shots before generating video, and beating baselines on recap quality and duration compliance.
Details
Novels have no direct audiovisual form short of a film adaptation, and film recaps depend on existing footage. The authors define a new task: generating a self-contained, source-grounded recap video straight from a novel within a fixed viewing time. ReVi is a dual-loop multi-agent framework built on edit first, generate second. Its inner loop plans story content, narration time and matching shots before any video is generated. An outer loop runs across novels and revises both the editing policy and the meta policy that governs those revisions, a form of recursive self-improvement. On two evaluation sets, ReVi improved recap quality and duration compliance over representative baselines, backed by human ratings and multi-round improvement results.