youtube-to-slides

name	youtube-to-slides
description	Convert a YouTube video into infographic slides. Extracts transcript, segments into sections, summarizes, and generates stylized infographic images using Gemini AI. 7 styles: davinci, magazine, comic, geek, chalkboard, collage, newspaper. Use when user wants slide summaries from YouTube.
argument-hint	<youtube-url> [--style davinci|magazine|comic|geek|chalkboard|collage|newspaper] [--max-sections N]
allowed-tools	Bash, Read, Write, Glob
context	fork

youtube-to-slides

Convert YouTube videos into beautiful infographic slide decks.

Natural Language Examples

Users can invoke this skill with natural language. Here are examples of what they might say and how to handle each:

Setup and troubleshooting:

• "Set up youtube-to-slides" → Run setup.sh
• "Check if youtube-to-slides is ready" → Run check-env.sh
• "I need to configure my API keys" → Guide to .env setup
• "It's not working" → Run check-env.sh, read diagnostics, consult references/TROUBLESHOOTING.md

Generate slides:

• "Generate slides from https://youtu.be/VIDEO_ID"
• "Convert this YouTube video to slides: https://youtube.com/watch?v=VIDEO_ID"
• "Make infographic slides from https://youtu.be/VIDEO_ID in comic style"
• "Create 5 magazine-style slides from https://youtu.be/VIDEO_ID"
• "Turn this video into slides https://youtu.be/VIDEO_ID --style geek"

Choose a style:

• "Use the davinci style" → --style davinci
• "Make it look like a magazine" → --style magazine
• "Comic book style please" → --style comic
• "Geek / bulletin board style" → --style geek
• "Chalkboard style" → --style chalkboard
• "Collage / dreamcore style" → --style collage
• "Newspaper style" → --style newspaper

Control slide count:

• "Just give me 4 slides" → --max-sections 4
• "Make as many slides as needed" → --max-sections 0

Preview without generating images:

• "Do a dry run first"
• "Just show me the prompts, don't generate images yet"

Argument Parsing

Parse $ARGUMENTS to extract:

• url (required) — YouTube video URL. Supports formats: https://youtu.be/ID, https://www.youtube.com/watch?v=ID, https://youtube.com/watch?v=ID
• --style (optional, default: davinci) — One of: davinci, magazine, comic, geek, chalkboard, collage, newspaper
• --max-sections (optional, default: 8) — Maximum number of slide sections to generate. Use 0 for unlimited.
• --dry-run (optional) — Show prompts without generating images
• --ar (optional, default: 16:9) — Aspect ratio: 16:9, 4:3, or 1:1

If no URL is provided, ask the user for one. If the URL looks invalid (not a YouTube URL), tell the user and ask for a valid one.

For style guidance, read references/STYLES.md in this skill directory.

Execution Steps

Step 1: Pre-flight Check

Run the environment check script:

bash "$SKILL_DIR/scripts/check-env.sh"

• If it exits 0, proceed to Step 2.
• If it exits non-zero, read its output for diagnostics:
  • Missing virtual environment or package — Run the setup script:

    bash "$SKILL_DIR/scripts/setup.sh"
    

    Then re-run check-env.sh to confirm.
  • Missing API key — Tell the user:
    1. Copy the example env file: cp "$SKILL_DIR/.env.example" "$SKILL_DIR/.env"
    2. Add their Gemini API key (get one free at https://aistudio.google.com/apikey)
    3. See references/TROUBLESHOOTING.md for detailed setup steps.
    • Note: The .env file is gitignored and stays local. The key is only sent to the Google Gemini API over HTTPS and is never logged or transmitted elsewhere.
  • After fixing, re-run check-env.sh to confirm everything passes.

If the user just asks to "set up" or "install" the skill, run both setup.sh and check-env.sh, then report the result.

Step 2: Run the Pipeline

Build the command from parsed arguments and run:

bash "$SKILL_DIR/scripts/run.sh" "<url>" --style <style> --max-sections <max_sections> --ar <ar>

Add --dry-run flag if requested.

This takes 3-5 minutes for a full run. Inform the user that generation is in progress and what style/settings are being used.

Step 3: Present Results

After successful completion:

1. Determine the video ID from the URL (the 11-character ID, e.g. twzLDx9iers from https://youtu.be/twzLDx9iers)
2. Read output/<video_id>/metadata.json to get the list of generated slides
3. Present a summary to the user:
   • Video title and channel
   • Style used
   • Number of slides generated
   • List each slide with its title and file path
4. Show the user how to open the slides:

   open output/<video_id>/
   

Error Handling

If the pipeline fails, check output for common errors and consult references/TROUBLESHOOTING.md:

• 429 Rate Limit — Gemini rate limit hit. The tool has built-in retry logic. If it still fails, suggest waiting 60 seconds and retrying.
• No transcript available — The video may not have captions. Inform the user.
• Invalid URL — Ask the user for a valid YouTube URL.
• API key errors — Guide user to set up their .env file with valid keys.

Security

• API key (Gemini only) is stored in a local .env file that is gitignored. It is only sent to the Google Gemini API over HTTPS for AI summarization and image generation. The key is never logged, printed, or transmitted to any other endpoint.
• Video metadata is fetched via yt-dlp, which scrapes public YouTube pages directly — no API key required.
• Scripts (setup.sh, check-env.sh, run.sh) are local shell scripts within this skill directory. They do not download or execute remote code. setup.sh creates a Python virtual environment and installs the local package. run.sh loads the .env file and invokes the yt-slides CLI. check-env.sh verifies the environment is configured.
• Network calls are limited to: Google Gemini API (AI summarization and image generation), YouTube (video metadata via yt-dlp, transcripts), and PyPI (only during setup.sh for dependency installation).

Variable Reference

• $SKILL_DIR — Absolute path to this skill's directory (skills/youtube-to-slides/)
• $ARGUMENTS — Raw argument string passed by the user