joeseesun GitHub avatar

qiaomu-cut-skill

joeseesun

qiaomu-cut is an agent-native skill that turns a one-sentence video brief into a reproducible, verifiable video project with governed media sourcing, multi-layer bilingual subtitles, and scalable rendering profiles.

Stars

288

7-day growth

+45

Forks

36

Open issues

0

License

MIT

Last updated

2026-07-19

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It bridges the gap between AI-generated video suggestions and actual production by providing a structured pipeline (QiaoCut IR, asset manifest, render profiles, cache) that agents can execute and verify, while integrating multiple real-world media sources (33台词, ClipSeek, ListenHub) with built-in licensing and security boundaries.

Who it is for

  • AI agent developers building video creation pipelines
  • Content creators who want automated, reproducible video production
  • Developers integrating media sourcing (dialogue, stock footage, TTS) into automated workflows
  • Teams needing verifiable video engineering with caching and multi-profile rendering

Use cases

  • Generate a 60-second English learning video from movie dialogue snippets
  • Create a product launch video with web animation overlays and TTS narration
  • Produce a bilingual explainer video using free stock footage and AI-generated images
  • Render a cinematic short film with custom intro/outro templates and three-level subtitles

Strengths

  • Provides a deterministic, source-aware pipeline (QiaoCut IR) that produces reproducible video projects
  • Supports three rendering profiles (preview, standard, final) with caching, allowing fast iteration before final output
  • Integrates managed media sourcing (33台词, ClipSeek, ListenHub) with provenance tracking, licensing reports, and security boundaries
  • Includes multi-layer bilingual ASS subtitles, 20 branded intro/outro templates, and built-in verification (doctor, render validation, licensing checks)

Considerations

  • Requires manual installation of dependencies (Node 18+, ffmpeg-full, ListenHub CLI) and platform-specific setup (macOS for full TTS and 33台词, Linux partial support)
  • Some advanced features (HTML video capture, Manim/PPT direct output, complex transitions) are not yet implemented and remain as extension points
  • Paid or remote operations require explicit user confirmation (--yes) and do not automatically handle licensing rights for sourced media; user must verify compliance

README quick start

qiaomu-cut

中文 | English

你只说一句“我要什么视频”,它把素材、分镜、字幕、转场、动效、渲染和质检组织成一个可复现的视频工程。 Say what video you want; qiaomu-cut turns it into a source-aware, renderer-ready, verifiable video project.

为什么值得用

普通 AI 会给你一段剪辑建议,真正开剪时还要你自己找素材、查版权、装 ffmpeg、写字幕、调转场、验编码。

qiaomu-cut 把这些拆成一个可复用的视频导演系统:它先生成 QiaoCut IR,再按环境路由到 33台词、ClipSeek/Pexels/Pixabay、本地素材、ListenHub/MarsWave 图片/视频/TTS/Voice/音乐、Coli 本地 ASR、agent 图片生成、HTML/Manim/PPT 等可用引擎,最终由 ffmpeg-full 合成。

v0.5 已真实实现的是:项目时间线渲染、三层双语 ASS、移动端大字号与平台安全区、20 套品牌片头片尾模板、无缩放快速片尾、完整句裁切规则、素材文件/场景去重门、macOS 与文件型旁白、程序音乐、原声混合、preview / standard / final 三档渲染、分层验收和项目内缓存;同时完整锁定 MarsWave 上游技能快照,并用安全 adapter 接入其当前 CLI 能力、付费/上传确认、私有任务 capture、生成物下载校验和 manifest 入库。HTML 视频捕获、Manim/PPT 直出、复杂遮罩、速度渐变和完整转场库目前仍是工作流接口与扩展方向,不冒充已经全部内置。

它不是承诺“魔法般永远一键完美”,而是把专业视频制作流程变成 agent 能执行、能验证、能继续扩展的工程。

一行安装

npx skills add joeseesun/qiaomu-cut-skill --skill qiaomu-cut

本地开发或手动安装:

mkdir -p ~/.agents/skills
cp -R qiaomu-cut ~/.agents/skills/qiaomu-cut

验证:

node ~/.agents/skills/qiaomu-cut/scripts/qcut.js doctor --json

你可以这样说

  • “搜索几句电影里的脏话台词,混剪成英语学习视频。”
  • “用免费素材做一个 60 秒挖掘机英语启蒙视频。”
  • “介绍乔布斯,做一个电影感人物短片,有时间线和档案照片。”
  • “把这个口播视频剪成小红书风格,强字幕、B-roll、卡点转场。”
  • “用 3Blue1Brown 风格解释 Transformer attention。”
  • “给我的网站做一个产品发布视频,网页动效融入剪辑。”
  • “用 AI 生成三个电影感转场镜头、中文旁白和配乐,再合成 45 秒预告片。”

它会做什么

  1. 把一句话需求变成 QiaoCut IR:时长、比例、受众、风格、分镜、素材策略、渲染器。
  2. 搜索、生成或整理素材:33台词、ClipSeek、Pexels/Pixabay 原站、本地文件、ListenHub 图片/视频/旁白/音乐、Coli ASR、网页/信息源。
  3. 设计专业镜头:推拉摇移、匹配剪辑、J/L cut、遮罩、字幕跟随、标题卡、动态图形。
  4. 选择渲染引擎:ffmpeg-full、HTML/HyperFrames-style、Motion/CSS/SVG、Manim、PPT/slide、composite。
  5. 先用 preview 快速迭代,内容锁定后再用 standardfinal 渲染。
  6. 输出与档位匹配的成片、contact sheet、缓存命中、阶段耗时和质检信息。

前置条件

  • Node.js 18+:node --version
  • 使用 ListenHub provider 时 Node.js 20+,并运行 scripts/bootstrap_listenhub.sh --check
  • macOS 推荐 Homebrew:brew --version
  • ffmpeg-full:scripts/bootstrap_macos.sh --check
  • 需要影视台词素材时,本机已安装并登录 33台词 App,并另行安装获得授权的 33tc CLI adapter;用 which 33tc 验证,或设置 QIAOMU_33TC_CLI
  • 需要发布到 GitHub 时,GitHub CLI 已登录:gh auth status
  • 需要 AI 图片生成时,当前 agent 环境提供图片生成工具。
  • OpenAPI 路由从环境变量 LISTENHUB_API_KEY 或 ListenHub 官方本机 credential store 读取凭据,命令使用 listenhub openapi ...;OAuth/internal 路由需用户明确执行 listenhub auth login,命令使用顶层 listenhub ...。两者不是同一认证命名空间,不要把 key 放进命令、项目或仓库;doct

Description

乔木智能视频导演 Skill:素材治理、双语字幕、品牌包装与可复现渲染 | Agent-native video director with governed sourcing and verifiable rendering

Related repositories

Similar projects matched by category, topics, and programming language.

0xsline
0xsline GitHub avatar

OpenChatCut

OpenChatCut is an open-source, local-first AI video editor with a real multitrack timeline, conversational agents, and MCP support, enabling AI-assisted editing that remains fully editable.

TypeScript
579
Alisa0808
Alisa0808 GitHub avatar

vox-director

An agent skill that fully automates the creation of Vox-style paper-collage explainer videos from a single topic using Atlas Cloud API and local ffmpeg.

Python
466
7-e1even
Featured
7-e1even GitHub avatar

learn-agent

A collection of notes on building coding agents, derived from the production development of the desktop agent Reina, with each mechanism simplified into a zero-dependency, single-file Node.js demo.

AI & Machine LearningLarge Language Models
218