Bggg Tiktok Readvideo
把 TikTok、Reels、YouTube Shorts、UGC 广告、本地 MP4/MOV/WebM 等视频拆成 Codex 可读的视频上下文。 当用户要求 Codex...
Install
npx skills add binggandata/bggg-skills --skill bggg-tiktok-readvideo -g -a claude-code -y- Repository
- binggandata/bggg-skills612
- Category
- Media
- Updated
- 3mo ago
- License
- MIT
- Context cost
- Mediumabout 1,505 tokens
- Runs scripts
- Yes, review before use
What it does
把 TikTok、Reels、YouTube Shorts、UGC 广告、本地 MP4/MOV/WebM 等视频拆成 Codex 可读的视频上下文。 当用户要求 Codex 看懂视频、读视频、分析视频、总结视频、找 hook、找爆点、提取字幕、 语音转文字、ASR、批量转写 TikTok 视频或音频、 生成视频 timeline、抽关键帧、做 contact sheet、分析口播/画面/节奏、制定 TikTok 9:16 剪辑方案、 或把原始视频素材变成 edit_plan.json 并用 FFmpeg 渲染短视频时，应该使用此 skill。 本 skill 独立内置 ffprobe/ffmpeg/whisper.cpp/tesseract 可选流程，不依赖其他 BGGG skills。
BGGG TikTok ReadVideo
这个 skill 的原则很简单：不要让 Codex 直接“看 mp4”。先把视频拆成可读、可搜索、可执行的上下文，再让 Codex 做判断、剪辑规划和渲染。
核心脚本：
python3 bggg-tiktok-readvideo/scripts/analyze_video.py "/path/to/input.mp4"
只需要转写音轨时使用独立转写脚本：
python3 bggg-tiktok-readvideo/scripts/transcribe_video.py "/path/to/video-or-folder" --recursive --srt
脚本会创建：
bggg-tiktok-readvideo/projects/YYYYMMDD_slug/
├── raw/input.mp4
├── analysis/
│ ├── metadata.json
│ ├── transcript.txt
│ ├── transcript.srt
│ ├── scenes.json
│ ├── timeline.md
│ ├── contact_sheet.jpg
│ ├── keyframes/
│ ├── audio_events.json
│ ├── ocr.json
│ └── analysis_manifest.json
└── output/
└── edit_plan.template.json
Default Workflow
- Run
scripts/analyze_video.pybefore making any claim about the video content. - If the user only asks for transcription/ASR, run
scripts/transcribe_video.pyinstead of the full visual analysis pipeline. - Read
analysis/timeline.md,analysis/scenes.json,analysis/transcript.srt,analysis/audio_events.json, andanalysis/ocr.json. - Inspect
analysis/contact_sheet.jpgas the visual overview. Use individualanalysis/keyframes/frame_XXXX.jpgwhen a scene needs closer visual reading. - If the user asks for TikTok editing, write
output/edit_plan.jsonbefore rendering. - Render with
scripts/render_tiktok.pyonly after the edit plan exists.
Do not infer content from the filename, title, or folder name alone. The timeline and contact sheet are the source of truth.
Analyze Commands
Basic local analysis:
python3 bggg-tiktok-readvideo/scripts/analyze_video.py "/path/to/video.mp4"
Give the run a stable project name:
python3 bggg-tiktok-readvideo/scripts/analyze_video.py "/path/to/video.mp4" \
--slug product_qc_ugc
Faster visual-only pass:
python3 bggg-tiktok-readvideo/scripts/analyze_video.py "/path/to/video.mp4" \
--no-transcribe --max-frames 24
Use a specific whisper.cpp model:
