rea-video Open-Source Tool Lets Agents Reverse-Engineer Any Video
The three-stage design keeps costs down by reserving large-model calls for interpretation rather than raw frame analysis, a pattern reusable in any video-understanding workflow. E-commerce and short-video teams that previously paid for viral-content replication services can now run the same pipeline themselves and fine-tune it to their own formats. It contrasts with paid video editing agent products that had bundled this capability as a selling point just months ago.
Developer @yao_june12106 released rea-video, an open-source project on GitHub that lets AI agents analyse, break down and replicate arbitrary videos. The pipeline starts with local tools that detect colour, pixel and brightness changes frame by frame to find cuts, then analyse motion, noise, repeated frames, text and audio without consuming model tokens. Multimodal models then explain each shot using low-resolution clips plus first, middle and last frames, outputting scene descriptions, production methods and replication prompts. Claude or Codex reviews the shot collages and findings to correct classifications and generate a final report.
[开源分享] rea-video 让 Agent 可以逆向、拆解和复刻任意视频，作者 @yao_june12106 之前合作的团队做 Video Editing Agent，其中也有「爆款复刻」这个环节，当时还想作为产品主打给电商场景的卖点，这才几个月，都有很多开源实现免费分享出来了。 github.com/JuneYaooo/rea-… # 一起看看逆向的重要流程和实际实现，再根据你的具体逆向场景，做微调就行 1. 先用本地工具拆解。 逐帧检测颜色、像素和亮度变化，识别切镜头；再分析运动、噪声、重复帧、文字和音频。本地规则先给出制作类型的初步判断及证据，这部分不消耗大模型 token。 2. 再让多模态模型解释。 默认按镜头分析；同一次拍摄被跳切分开的片段，可以合并请求。每次通常发送低分辨率视频片段、首尾帧和中间帧拼图，并附上本地特征摘要，让模型输出画面描述、制作方式及复刻提示词。默认还会单独分析一次带音频的整片，理解整体结构和声音。 3. 最后由 Claude／Codex 复核。 Agent 查看镜头拼图和分析结果，修正分类、补充依据，再生成报告。因此最终结论来自“本地测量＋模型分析＋Agent 复核”，而不直接恢复原始工程。 💬 7 🔄 0 ❤️ 5 👀 439 📊 8 ⚡
