发表机构
School of Artificial Intelligence, The Chinese University of Hong Kong, Shenzhen; vivo AI Lab; University of the Chinese Academy of Sciences; Southeast University; Peking University; School of Science and Engineering, The Chinese University of Hong Kong, Shenzhen; Shenzhen Future Network of Intelligence Institute (FNii-Shenzhen); Guangdong Provincial Key Laboratory of Future Networks of Intelligence, CUHK-Shenzhen（香港中文大学（深圳）人工智能学院; vivo AI实验室; 中国科学院大学; 东南大学; 北京大学; 香港中文大学（深圳）理工学院; 深圳未来智能网络研究院; 香港中文大学（深圳）广东省未来智能网络重点实验室）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对可执行视频编辑规划问题，提出基于验证器的RefineCut及RefineCut-Evo方法，训练8B开放权重规划器，在基准上取得优于前沿模型的效果，代码与基准已公开。
AI 中文摘要
实用视频编辑不仅是像素生成：编辑者必须将脚本、素材池、音乐元数据和硬约束转化为可执行时间线。我们将该决策层研究为「可执行视频编辑规划」，并推出RefineCut，与包装提示前沿模型的工作流系统不同，它为此训练了一个紧凑的开放权重规划器。该规划器通过结构化补丁编辑类型化时间线，涵盖素材选择、修剪、排序、转场、时长及音乐对齐；确定性验证器应用每个补丁并对照显式约束账本检查。由于编辑无单一真实修复，我们不直接模仿教师：RefineCut通过验证器重放所有多教师分支，保留验证器最优修复作为监督。第二阶段RefineCut-Evo让学生用验证器和任务准则评分自身修复，并基于高边际偏好对训练，最终8B规划器在推理时以封闭验证器循环运行，无需教师调用。在RefineCut-Bench（3578个任务、7971个带字幕素材、499条音乐曲目、显式账本）上，验证器重放蒸馏使规划器在协议特定视频编辑得分从0.620提升至0.858，RefineCut-Evo达到0.924；该增益可迁移至Llama-3.1-8B和GLM-4-9B，在同一封闭循环中，8B规划器表现匹配或超越其前沿教师。代码和RefineCut-Bench已公开，详见数据可用性声明。
英文摘要
Practical video editing is not only pixel generation: an editor must turn a brief, a clip pool, music metadata, and hard constraints into an executable timeline. We study this decision layer as \emph{executable video-editing planning} and introduce RefineCut, which, unlike workflow systems that wrap a prompted frontier model, trains a compact open-weight planner for it. The planner edits a typed timeline through structured patches covering clip selection, trimming, ordering, transitions, and duration and music alignment; a deterministic verifier applies each patch and checks it against an explicit constraint ledger. Because editing has no single ground-truth repair, we do not imitate teachers directly: RefineCut replays every multi-teacher branch through the verifier and keeps verifier-best repairs as supervision. A second stage, RefineCut-Evo, lets the student score its own repairs with the verifier and a task rubric and trains on high-margin preference pairs, so the final $8$B planner runs in a closed verifier loop with no teacher calls at inference. On RefineCut-Bench ($3{,}578$ tasks, $7{,}971$ captioned clips, $499$ music tracks, explicit ledgers), verifier-replayed distillation lifts the planner from $0.620$ to $0.858$ on the protocol-specific Video-Editing Score and RefineCut-Evo reaches $0.924$; the gain transfers to Llama-3.1-8B and GLM-4-9B, and in the same closed loop the $8$B planner matches or exceeds its frontier teachers. Code and RefineCut-Bench are publicly released; see the Data Availability statement.
CommentsAccepted to the Main Conference of EMNLP '26
