videocut

分类: 内容与多媒体 | 上传者: zinan92 | 下载: 0 | 版本: v1.0（最新）

人工智能驱动的视频编辑——转录、自动剪切、字幕、挂钩、剪辑、封面、速度、管道。阅读本文以了解何时以及如何使用视频剪辑。

更新日志: Source: GitHub https://github.com/zinan92/videocut

目录结构

当前层级: tree/main/

📁 capabilities/
- 📁 autocut/
  - 📁 rules/
    
    📄 1-核心原则.md 435 B
    
    📄 2-语气词检测.md 645 B
    
    📄 3-静音段处理.md 876 B
    
    📄 4-重复句检测.md 3.1 KB
    
    📄 5-卡顿词.md 514 B
    
    📄 6-句内重复检测.md 772 B
    
    📄 7-连续语气词.md 559 B
    
    📄 8-重说纠正.md 949 B
    
    📄 9-残句检测.md 2.1 KB
    
    📄 feedback_examples.md 447 B
    
    📄 README.md 1.3 KB
  - 📄 cut.sh 3.9 KB
  - 📄 feedback_aggregator.js 3.7 KB
  - 📄 index.js 6.6 KB
  - 📄 review_server.js 14.0 KB
  - 📄 SKILL.md 4.0 KB
- 📁 clip/
  - 📄 index.js 6.3 KB
  - 📄 SKILL.md 1.5 KB
  - 📄 split.sh 475 B
- 📁 cover/
  - 📄 generate.sh 6.1 KB
  - 📄 index.js 2.9 KB
  - 📄 SKILL.md 551 B
- 📁 hook/
  - 📄 index.js 7.4 KB
  - 📄 match.js 7.2 KB
  - 📄 SKILL.md 2.2 KB
- 📁 speed/
  - 📄 adjust.sh 513 B
  - 📄 index.js 1.5 KB
  - 📄 SKILL.md 437 B
- 📁 subtitle/
  - 📄 burn.sh 609 B
  - 📄 detect.js 3.1 KB
  - 📄 index.js 2.8 KB
  - 📄 SKILL.md 3.1 KB
- 📁 transcribe/
  - 📄 generate_words.js 3.2 KB
  - 📄 index.js 7.0 KB
  - 📄 mlx_transcribe.py 694 B
  - 📄 SKILL.md 721 B
  - 📄 whisper.sh 3.4 KB
📁 docs/
- 📁 superpowers/
  - 📁 plans/
    
    📄 2026-03-18-parallel-retry-feedback.md 20.1 KB
    
    📄 2026-03-18-publish-sh.md 24.2 KB
    
    📄 2026-03-18-web-dashboard.md 36.0 KB
    
    📄 2026-03-19-video-split.md 768 B
    
    📄 2026-03-30-videocut-modularization.md 34.2 KB
  - 📁 specs/
    
    📄 2026-03-19-video-split-design.md 1.8 KB
    
    📄 2026-03-30-videocut-modularization-design.md 14.7 KB
📁 examples/
- 📄 4_article_cn.md 2.9 KB
- 📄 4_card_1.png 134.6 KB
- 📄 4_thumbnail.png 477.2 KB
- 📄 5_jike_post.md 1.8 KB
- 📄 5_x_post.md 259 B
- 📄 5_xhs_caption.md 1.6 KB
📁 lib/
- 📄 claude.js 2.9 KB
- 📄 ffmpeg.js 2.5 KB
- 📄 srt.js 5.8 KB
📁 references/
- 📄 README.md 1.6 KB
📁 tests/
- 📁 capabilities/
  - 📄 autocut.test.js 3.5 KB
  - 📄 clip.test.js 3.4 KB
  - 📄 cover.test.js 264 B
  - 📄 hook.test.js 2.1 KB
  - 📄 speed.test.js 442 B
  - 📄 subtitle.test.js 612 B
  - 📄 transcribe-mlx.test.js 1.3 KB
  - 📄 transcribe.test.js 2.7 KB
- 📁 lib/
  - 📄 claude.test.js 848 B
  - 📄 ffmpeg.test.js 4.2 KB
  - 📄 srt.test.js 11.7 KB
- 📄 cli.test.js 3.7 KB
- 📄 integration.test.js 3.6 KB
- 📄 pipeline.test.js 499 B
- 📄 test_e2e_split.js 5.8 KB
- 📄 test_e2e_split.sh 4.6 KB
- 📄 test_e2e_verify.sh 4.1 KB
- 📄 test_feedback_aggregator.js 3.6 KB
- 📄 test_generate_srt.js 3.2 KB
📁 web/
- 📁 src/
  - 📁 components/
    
    📄 ChapterCard.tsx 1.5 KB
    
    📄 ChapterTimeline.tsx 2.0 KB
    
    📄 FileUpload.tsx 2.1 KB
    
    📄 PhaseProgress.tsx 2.0 KB
    
    📄 PlatformCard.tsx 2.8 KB
  - 📁 lib/
    
    📄 api.ts 1.3 KB
  - 📁 pages/
    
    📄 PipelinePage.tsx 5.5 KB
    
    📄 PublishPage.tsx 5.8 KB
    
    📄 SplitPage.tsx 9.6 KB
  - 📄 App.tsx 2.1 KB
  - 📄 index.css 2.9 KB
  - 📄 main.tsx 317 B
- 📄 .gitignore 27 B
- 📄 index.html 661 B
- 📄 package-lock.json 135.2 KB
- 📄 package.json 808 B
- 📄 postcss.config.js 80 B
- 📄 server.js 17.6 KB
- 📄 tailwind.config.js 578 B
- 📄 tsconfig.json 555 B
- 📄 vite.config.ts 280 B
📁 剪口播/
- 📁 scripts/
  - 📄 cut_video.sh 3.9 KB
  - 📄 detect_hardcoded_subtitles.js 3.1 KB
  - 📄 feedback_aggregator.js 3.7 KB
  - 📄 generate_review.js 17.6 KB
  - 📄 generate_srt.js 2.7 KB
  - 📄 generate_subtitles.js 3.2 KB
  - 📄 review_server.js 14.0 KB
  - 📄 volcengine_transcribe.sh 2.8 KB
  - 📄 whisper_transcribe.sh 2.4 KB
- 📁 用户习惯/
  - 📄 1-核心原则.md 435 B
  - 📄 2-语气词检测.md 645 B
  - 📄 3-静音段处理.md 876 B
  - 📄 4-重复句检测.md 3.1 KB
  - 📄 5-卡顿词.md 514 B
  - 📄 6-句内重复检测.md 772 B
  - 📄 7-连续语气词.md 559 B
  - 📄 8-重说纠正.md 949 B
  - 📄 9-残句检测.md 2.1 KB
  - 📄 README.md 1.3 KB
- 📄 SKILL.md 6.8 KB
📁 字幕/
- 📁 scripts/
  - 📄 subtitle_server.js 21.9 KB
- 📄 README.md 366 B
- 📄 SKILL.md 5.3 KB
- 📄 词典.txt 110 B
📁 安装/
- 📄 README.md 311 B
- 📄 SKILL.md 1.9 KB
📁 自进化/
- 📄 README.md 484 B
- 📄 SKILL.md 2.7 KB
📄 .gitignore 71 B
📄 capability.json 3.4 KB
📄 cli.js 4.9 KB
📄 content-repurpose.sh 15.7 KB
📄 generate-cards.sh 6.1 KB
📄 package.json 245 B
📄 pipeline.js 2.5 KB
📄 pipeline.sh 24.3 KB
📄 PIPELINE_SPEC.md 2.9 KB
📄 publish.sh 15.6 KB
📄 README.md 13.8 KB
📄 run.sh 10.6 KB
📄 SKILL.md 5.2 KB

SKILL.md

---
name: videocut
description: AI-powered video editing — transcribe, autocut, subtitle, hook, clip, cover, speed, pipeline. Read this to know WHEN and HOW to use videocut.
---

# Videocut

AI-powered video editing capabilities for spoken-word content. Seven independent tools plus a pipeline mode for chaining them.

## When to Use

Use `videocut` when the user has a **video file** and wants to **edit, transcribe, or produce content from it**.

| User says | Action |
|-----------|--------|
| "转录 / transcribe this video" | `videocut transcribe input.mp4 -o output/` |
| "去口癖 / 去废话 / remove fillers" | `videocut autocut input.mp4 -o output/` |
| "加字幕 / burn subtitles" | `videocut subtitle input.mp4 -o output/` |
| "提取金句 / find hooks" | `videocut hook input.mp4 -o output/` |
| "拆成短视频 / split into clips" | `videocut clip input.mp4 -o output/` |
| "做封面 / generate quote cards" | `videocut cover -o output/ --text "金句"` |
| "加速 / speed up" | `videocut speed input.mp4 -o output/ --rate 1.1` |
| "一条龙 / full production" | `videocut pipeline input.mp4 --steps autocut,subtitle,hook -o output/` |

## When NOT to Use

| User wants | Use instead |
|------------|-------------|
| Download a video from URL | `content-downloader` |
| Extract text from downloaded content (structured) | `content-extractor` |
| Rewrite content for another platform | `content-rewriter` |
| Analyze content trends | `content-intelligence` |

## Sub-Capabilities

### transcribe
Transcribe video/audio to text with word-level timestamps.
- **Input:** Video/audio file
- **Output:** `transcript.json` (word-level), `transcript.txt`, `transcript.srt`
- **AI:** Whisper only (no Claude)

### autocut
Remove filler words, stutters, and dead air from spoken-word video.
- **Input:** Video file
- **Output:** `cut.mp4`, `delete_segments.json`, `cut_feedback.json`
- **AI:** Whisper + Claude (9 extensible rule files for analysis)

### subtitle
Detect existing hardcoded subtitles or generate and burn new ones.
- **Input:** Video file
- **Output:** `subtitled.mp4`, `subtitle.srt`
- **AI:** Whisper for transcription

### hook
Extract the most engaging quotes and cut corresponding video segments.
- **Input:** Video file
- **Output:** `hook.mp4` (concatenated), `hooks.json`, `hook_segments/*.mp4`
- **AI:** Claude selects quotes, character-level text matching

### clip
Split long video into chapters based on topic changes.
- **Input:** Video file
- **Output:** `clips/*.mp4`, `chapters.json`
- **AI:** Claude detects chapter boundaries

### cover
Generate quote card images from hooks or custom text.
- **Input:** `hooks.json` or `--text` flag
- **Output:** `card_1.png ... card_N.png` (1080x1080)
- **AI:** None (Chrome Headless rendering)

### speed
Speed up video with pitch-preserving audio adjustment.
- **Input:** Video file + rate (1.0–1.2x recommended)
- **Output:** `speed.mp4`
- **AI:** None (FFmpeg atempo filter)

## Pipeline Mode

Chain multiple sub-capabilities in sequence:

```bash
videocut pipeline input.mp4 --steps autocut,speed,subtitle,hook,cover -o output/
```

### Recommended Order

```
autocut → speed → subtitle → hook → clip → cover
```

- **autocut** first: cut filler before anything else
- **speed** before subtitle: subtitles generated against final-speed video
- **subtitle** before hook: hook can reuse the SRT file
- **cover** last: needs hooks.json from hook step

## CLI Reference

```bash
videocut <capability> <input> [OPTIONS]
```

### Global Flags

| Flag | Default | Description |
|------|---------|-------------|
| `-o` / `--output` | `./output/{date}_{basename}/` | Output directory |
| `--no-review` | False | Skip interactive review (autocut) |

### Capability-Specific Flags

| Capability | Flag | Default | Description |
|-----------|------|---------|-------------|
| hook | `--count` | 5 | Number of hooks to extract |
| speed | `--rate` | 1.1 | Speed multiplier (1.0–1.2) |
| pipeline | `--steps` | required | Comma-separated capability list |
| autocut | `--model` | — | Claude model override |

## Architecture

```
cli.js (argument parser + router)
  → capabilities/{name}/index.js (per-capability orchestrator)
    → lib/ffmpeg.js (video probing, audio extraction)
    → lib/srt.js (SRT parsing, generation, merging)
    → lib/claude.js (Claude CLI wrapper with retry)
    → capabilities/{name}/*.sh (FFmpeg operations)
```

Each sub-capability is a self-contained directory with its own `index.js`, bash scripts, and `SKILL.md`.

## Dependencies

- Node.js 18+
- FFmpeg (with libass for subtitle burning)
- Whisper (`pip install openai-whisper`)
- Claude CLI (authenticated, on PATH)
- Chrome/Chromium (for cover generation only)

## Failure Modes

| Failure | Behavior |
|---------|----------|
| Input file not found | Exit 1 with error message |
| Unknown capability | Exit 1 with supported capability list |
| FFmpeg error | Exit with FFmpeg stderr |
| Claude API error | Retry with exponential backoff (1s, 3s, 9s) |
| Whisper failure | Exit with transcription error |

## Pipeline Position

```
content-downloader → videocut (standalone video editing)
                   → content-extractor → content-rewriter
```

Videocut operates independently on video files. It does not require content-downloader output — any video file works.

登录后下载/点赞/收藏 ❤ 8 | ★ 0

videocut

目录结构

SKILL.md

举报内容

提示