Zum Inhalt springen
TakoAPI
Zurück zu den Skills

video-understanding

von bill492

Coding Agents & IDEs
605Downloadzahlen laut clawskills.sh
4Sterne
0Aufrufe
0 LikesMelde dich an, um diesen Skill zu liken
—Noch nicht bewertet
Deine Bewertung

Kurzbeschreibung

Analyze videos with Google Gemini multimodal AI.

Beschreibung

Was dieser Skill macht

Analyzes videos using Google Gemini's multimodal AI. Downloads from 1000+ sources including YouTube, Loom, TikTok, and Vimeo via yt-dlp. Returns transcripts, visual descriptions, summaries, and answers to specific questions about video content.

YouTube URLs go directly to Gemini without a download step, and all other platforms are handled automatically via yt-dlp, so no manual downloading or conversion is needed.

Wann du ihn einsetzt

  • Transcribing a Loom screen recording into meeting notes
  • Summarizing a YouTube tutorial without watching the full video
  • Extracting steps from a product demo or sales recording
  • Answering a specific question about a TikTok or Instagram video
  • Archiving a Twitter/X clip with transcript and visual description

Beispiel-Workflow

EINGABE

User asks: What decisions were made in this Loom recording? https://www.loom.com/share/abc123

AGENT
  1. 1

    Detects a non-YouTube URL and runs yt-dlp to download the video file

  2. 2

    Uploads the file to the Gemini File API and polls until processing completes

  3. 3

    Sends the structured analysis prompt plus the user question to Gemini

  4. 4

    Cleans up the temp file and Gemini upload after receiving the response

AUSGABE

JSON with timestamped transcript, visual description, 2-3 sentence summary, and a targeted answer field addressing the question about decisions made

Voraussetzungen

  • GEMINI_API_KEY environment variable from Google AI Studio
  • yt-dlp installed (brew install yt-dlp or pip install yt-dlp)
  • ffmpeg installed (brew install ffmpeg)

Installation

clawhub install bill492/video-understanding

Führe den CLI-Befehl aus oder füge den Prompt in Claude Code, Cursor, Codex oder deinen Coding-Agenten ein.

Links