modbender video-intel
Downloads videos and extracts transcripts using yt-dlp (captions) with OpenAI Whisper fallback.
Required Dependencies
| Dependency | Purpose | Required? |
|---|---|---|
yt-dlp |
Fetch captions and download audio/video | ✅ Always |
python3 |
Parse VTT/SRT caption files | ✅ Always |
curl |
Call OpenAI Whisper API | ✅ For Whisper fallback |
ffmpeg |
Extract audio from TikTok/Instagram/X | ⚠️ Non-YouTube only |
OPENAI_API_KEY |
Authenticate with OpenAI Whisper API | ⚠️ Only if captions unavailable |
Install binaries:
# yt-dlp
curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp -o ~/bin/yt-dlp && chmod +x ~/bin/yt-dlp
# ffmpeg (Debian/Ubuntu)
sudo apt install ffmpeg
⚠️ Privacy Notice — OpenAI Audio Upload
When captions are unavailable (common for TikTok, Instagram, X), the script downloads the audio and uploads it to OpenAI's transcription API (https://api.openai.com/v1/audio/transcriptions). This means:
- Audio content leaves your machine and is sent to OpenAI
- Requires
OPENAI_API_KEYto be set - If you don't want external transmission: don't set
OPENAI_API_KEY, or use a local transcription model
YouTube videos almost always have captions and will not trigger an upload.
Script
~/.openclaw/skills/video-intel/scripts/video-intel.sh
Workflows
Get transcript
~/.openclaw/skills/video-intel/scripts/video-intel.sh transcript <url>
- YouTube: uses built-in captions/auto-subs (fast, no audio download or external upload)
- TikTok/Instagram/X: downloads audio → uploads to OpenAI Whisper for transcription
- Preferred language:
--lang ptfor Portuguese
Get video info
~/.openclaw/skills/video-intel/scripts/video-intel.sh info <url>
List available caption tracks
~/.openclaw/skills/video-intel/scripts/video-intel.sh captions <url>
Download video
~/.openclaw/skills/video-intel/scripts/video-intel.sh download <url> [--format audio|720p|best]
After getting transcript
- Summary: Summarize in 3-5 bullet points
- Key quotes: Extract most notable quotes
- Full summary: Write a paragraph summary with context
- Translation: Translate to the user's language if different
Notes
- YouTube auto-captions are usually available even without ffmpeg
- TikTok/Instagram/X require ffmpeg for audio extraction
- Large videos (>25MB audio) may hit OpenAI's file size limit — use
--format audioto get a smaller mp3 - Output cached in
/tmp/video-intel/by default