LeoYeAI/agent-master-skills / skills/auto-video-editing
video-editing
VerifiedAutomated video editing skill for talk/vlog/standup videos. Use when: cutting video, splitting video into sentences, merging video clips, extracting audio, transcribing speech, auto-editing oral presentation videos, combining selected sentence clips into a final video, generating video cover/thumbnail with title. Requires ffmpeg and whisper.
| File | Kind | Size |
|---|---|---|
| README.md | markdown | 6.8 KB |
| SKILL.md | skill md | 16.4 KB |
| _meta.json | config | 290 B |
| scripts/add_chapter_bar.py | script | 11.5 KB |
| scripts/burn_subtitles.py | script | 9.7 KB |
| scripts/extract_audio.py | script | 1.4 KB |
| scripts/extract_audio.sh | script | 742 B |
| scripts/generate_cover.py | script | 9.9 KB |
| scripts/merge_clips.py | script | 5.3 KB |
| scripts/split_video.py | script | 3.6 KB |
| scripts/transcribe.py | script | 5.4 KB |
| scripts/utils.py | script | 18.0 KB |
README.md
View on GitHub# Video Editing Skill for OpenClaw
一个基于 AI 的自动视频剪辑 Skill,专为口播/脱口秀/Vlog 类视频设计。
它可以自动将你的视频按每句话切分,去掉口误和卡壳,加上字幕,最终合成一个可以直接发布的短视频。
> 适用于:小红书、抖音、视频号等竖屏短视频平台
## 功能特性
- **语音识别切分** — 支持 faster-whisper(推荐,4x 加速)和 openai-whisper,自动按句子切分
- **GPU 硬件加速** — 自动检测 NVIDIA NVENC / Apple VideoToolbox / Intel QSV / AMD AMF
- **精确剪辑** — 重编码模式切割,音频在句子边界精确切断,无重复尾音
- **自动字幕** — 中英文自动检测,自动折行,竖屏位置优化
- **灵活合成** — 选择需要的片段,按任意顺序合成最终视频
- **自动封面** — 将视频第一帧替换为带标题的封面,自动根据内容总结标题
- **章节时间轴** — 自动分章,在视频底部/顶部叠加彩色章节进度条,横屏竖屏自适应
- **多视频支持** — 同时处理多个视频文件,跨视频混合选择片段
- **跨平台** — 支持 macOS / Linux / WSL / Windows
- **中国加速** — 自动检测中国区域,使用清华 pip 镜像和 HuggingFace 镜像
## 安装
### 方式一:通过 OpenClaw Skills 安装(推荐)
```bash
openclaw skills add https://github.com/maxazure/video-editing-skill.git
```
或手动克隆到 OpenClaw skills 目录:
```bash
git clone https://github.com/maxazure/video-editing-skill.git ~/.openclaw/skills/video-editing
```
### 方式二:手动克隆
```bash
git clone https://github.com/maxazure/video-editing-skill.git
cd video-editing-skill
```
### 安装系统依赖
**macOS:**
```bash
brew install ffmpeg
```
**Ubuntu/Debian/WSL:**
```bash
sudo apt install ffmpeg fonts-noto-cjk
```
**Windows:**
建议使用 WSL2 环境。在 WSL 中按 Ubuntu 方式安装即可。
### 安装 Python 依赖
```bash
python3 -m venv .venv
source .venv/bin/activate # Windows WSL 同样使用此命令
pip install faster-whisper # 推荐,速度快 4 倍
```
中国用户加速安装:
```bash
pip install faster-whisper -i https://pypi.tuna.tsinghua.edu.cn/simple
```
> Whisper 模型首次运行时会自动下载。中国用户会自动使用 HuggingFace 镜像 (hf-mirror.com)。
### 验证安装
```bash
python3 scripts/utils.py # 运行环境诊断
```
这会输出完整的环境报告:平台、GPU、编码器、Whisper 引擎、推荐模型等。
## 使用方式
### 配合 OpenClaw / Claude Code 使用(推荐)
安装完依赖后,用自然语言告诉 AI:
```
帮我剪一下 videos/ 目录下的视频,按句子切分,去掉口误,加上字幕
```
AI 会自动调用各个脚本,完成整个剪辑流程。
### 手动使用
#### Step 1: 提取音频
```bash
python3 scripts/extract_audio.py "your_video.mp4"
# 输出: your_video_audio.wav
```
#### Step 2: 语音识别
```bash
source .venv/bin/activate
python3 scripts/transcribe.py "your_video_audio.wav" --model auto --language zh
# 输出: your_video_transcript.json
```
`--model auto` 会根据硬件自动选择最佳模型:
| 硬件 | 自动选择 | 原因 |
|------|---------|------|
| NVIDIA GPU | large-v3 | CUDA 加速,大模型也很快 |
| Apple Silicon | large-v3-turbo | 速度与质量的最佳平衡 |
| Intel/AMD 集显 | medium | CPU+iGPU 的最佳平衡 |
| 纯 CPU | small | 速度优先 |
也可手动指定:`--model large-v3` / `--model medium` 等
#### Step 3: 视频切分
```bash
python3 scripts/split_video.py "your_video.mp4" "your_video_transcript.json"
# 输出: your_video_clips/ 目录
```
#### Step 4: 烧录字幕
```bash
python3 scripts/burn_subtitles.py "your_video_clips" "your_video_transcript.json"
# 输出: your_video_clips_subtitled/ 目录
```
#### Step 5: 合成视频
```bash
python3 scripts/merge_clips.py "your_video_clips_subtitled" --select "1-5,8,10-15" --output "final.mp4"
```
#### Step 6: 生成封面
```bash
python3 scripts/generate_cover.py "final.mp4" --title "你的封面标题" --transcript "your_video_transcript.json"
# 输出: final_with_cover.mp4
```
#### Step 7: 添加章节时间轴
```bash
python3 scripts/add_chapter_bar.py "final.mp4" --transcript "your_video_transcript.json"
# 输出: final_chapters.mp4
```
自动根据转录内容分章,在视频上叠加彩色章节进度条。横屏视频放在底部,竖屏视频放在顶部(避开抖音/小红书底部 UI)。
可选参数:
- `--style mono` — 单色灰白风格(默认 `color` 彩色)
- `--chapters chapters.json` — 自定义章节 JSON
#### Step 8(可选): 调整播放速度
```bash
ffmpeg -i final.mp4 \
-filter_complex "[0:v]setpts=PTS/1.25[v];[0:a]atempo=1.25[a]" \
-map "[v]" -map "[a]" \
-c:v libx264 -preset fast -crf 18 -c:a aac -b:a 192k \
final_1.25x.mp4
```
## 目录结构
```
video-editing-skill/
├── README.md # 本文件
├── SKILL.md # Skill 定义(OpenClaw 读取)
├── scripts/
│ ├── utils.py # 共享工具(平台/GPU/字体/镜像检测)
│ ├── extract_audio.py # 音频提取
│ ├── extract_audio.sh # 音频提取(bash 版,兼容旧流程)
│ ├── transcribe.py # 语音识别(faster-whisper / openai-whisper)
│ ├── split_video.py # 视频切分
│ ├── burn_subtitles.py # 字幕烧录
│ ├── merge_clips.py # 视频合成
│ ├── generate_cover.py # 封面生成
│ └── add_chapter_bar.py # 章节时间轴
├── fonts/ # 字体缓存(自动下载)
└── videos/ # 用户视频工作目录
```
## 技术细节
| 组件 | 技术 |
|------|------|
| 语音识别 | faster-whisper (CTranslate2) / OpenAI Whisper |
| 视频编码 | NVENC / VideoToolbox / QSV / AMF / libx264(自动检测) |
| 字幕渲染 | ASS 格式 + Noto Sans SC / PingFang SC / Microsoft YaHei |
| 视频切分 | 精确重编码(非 stream copy) |
| 平台检测 | macOS / Linux / WSL / Windows 自动识别 |
### 硬件加速编码器优先级
```
NVIDIA NVENC > Apple VideoToolbox > Intel QSV > AMD AMF > CPU libx264
```
### 字幕字体优先级
```
自定义字体 > Noto Sans SC (自动下载) > PingFang SC (macOS) > Microsoft YaHei (Windows/WSL) > fc-match
```
中国用户字体下载使用 jsDelivr CDN 加速,无需访问 GitHub。
### 中国用户优化
- pip 安装自动使用清华镜像
- Whisper 模型自动使用 hf-mirror.com 下载
- 字体下载使用 jsDelivr CDN 备用源
- 可通过 `--mirror` 参数或 `USE_CN_MIRROR=1` 环境变量强制启用
## 系统要求
- macOS / Linux / WSL / Windows
- Python 3.8+
- FFmpeg(需包含 libass、libfreetype)
- 磁盘空间:约 1-3GB(取决于 Whisper 模型大小)
## License
MIT
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |