WeVisCut coordinates narration to long videos with efficient editing agent
WeVisCut: The Whole Editing Room in One Agent From Narration to Long-Form Video
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Making long videos from scripts is hard because many decisions must fit together well. The authors created WeVisCut, a system that learns to edit videos from narration by observing how professional videos were made and using those lessons to guide its choices. It trains a single smart agent that keeps track of editing steps, fixes mistakes, and works faster than big general models. They tested it on a large set of videos and a new benchmark, showing it works well and is more efficient at producing long narrated videos.
What this means in practice
- •For video production teams: Automate editing of long narrated videos with a single learned agent that efficiently manages complex editing decisions and recovers from errors.
- •For content platform operators: Improve automated video assembly systems that produce coherent long-form content from scripts reducing computational resource use while maintaining quality.
Authors
Chenyu Mu, Wenming Weng, Qinrui Wang, Xingsong Ye, Hao Yu, Jun He, Chong Sun, Chen Li, Jing Lyu, Erkun Yang, Cheng Deng
Abstract
Producing long-form videos from narration requires interdependent decisions that remain coherent across shots. Existing agentic editors often use multi-stage workflows driven by general models, leaving the quality-cost trade-off of compact, task-adapted policies unclear. We address this gap with WeVisCut, a long-form script-to-video system that connects tool execution with policy learning. Its agent harness maintains shared production state so that tool feedback informs subsequent decisions and repairs. Because finished videos do not expose these decisions, our four-stage data pipeline combines editing inversion with verified replay to turn recovered editing intent into executable supervision. Applied to 3,750 finished videos, this pipeline yields over 130K training examples. We use this supervision to train a single 9B policy that coordinates production and recovers from failures within the same harness. Moreover, we introduce a 100-task benchmark and five metrics assessing editing quality and delivery reliability. Under a shared tool environment, WeVisCut substantially outperforms its base model and achieves comparable performance to Claude Opus 5 while using nearly 33x fewer generated tokens per successful output minute. It also outperforms task-matched native editors on our proposed metrics.