razbakov/video-podcast-producer

Video Podcast Producer

Overview

Transform raw video/audio recordings into publish-ready content: trimmed, split by topic, with thumbnails and YouTube metadata. Uses ffmpeg for video processing and the brand-poster skill for professional thumbnails.

Prerequisites

• ffmpeg (for audio extraction and video cutting)

Workflow

Step 1: Transcribe

Use the transcribe-via-faster-whisper skill to transcribe the recording. This produces transcription.json (timestamped segments) and transcription.txt (full text).

Step 2: Analyze and Plan Cuts

Review timestamped segments to identify:

• Topic boundaries — where conversation shifts to a new subject
• Dead time to remove — login screens, tech issues, breaks, off-topic tangents, silence
• Best 60s hooks — most engaging moments for Shorts

Present a cut plan table to the user: | # | Title idea | Time range | Duration | |---|-----------|------------|----------|

Wait for user approval before cutting.

Step 3: Cut Video Clips

Use ffmpeg stream copy for speed (no re-encoding):

# Single segment
ffmpeg -y -i "$SRC" -ss START -to END -c copy output.mp4

# Multi-part clip (skip dead time within a topic)
ffmpeg -y -i "$SRC" -ss START1 -to END1 -c copy /tmp/partA.mp4
ffmpeg -y -i "$SRC" -ss START2 -to END2 -c copy /tmp/partB.mp4
echo "file '/tmp/partA.mp4'
file '/tmp/partB.mp4'" > /tmp/concat.txt
ffmpeg -y -f concat -safe 0 -i /tmp/concat.txt -c copy output.mp4

Run independent cuts in parallel with & and wait.

Step 4: Cut Shorts (9:16 Vertical)

For each clip, pick the most engaging 50-70 second moment. Crop center and scale to 1080x1920:

ffmpeg -y -ss START -to END -i "$SRC" \
  -vf "crop=405:720:437:0,scale=1080:1920" \
  -c:v libx264 -crf 23 -c:a aac output-short.mp4

The crop values assume 1280x720 source. Adjust for other resolutions:

• Crop width = source_height * 9/16
• Crop x offset = (source_width - crop_width) / 2

Step 5: Generate Thumbnails

Use the brand-poster skill to generate thumbnails for each clip.

For each clip, generate two thumbnails:

• 16:9 landscape (1280x720) — YouTube video thumbnail
• 9:16 vertical (1080x1920) — Shorts thumbnail

Design principles:

• Massive bold text filling the frame — readable at small sizes
• 2-3 color palette per thumbnail (gradient background + accent)
• Clean geometric elements, no clutter
• Strong visual hierarchy: title > graphic > subtitle

Step 6: Generate YouTube Metadata

For each clip, generate:

• Title — under 70 chars, hook-driven, not clickbait
• Description — summary + topic bullets + relevant links + hashtags
• Tags — 8-12 relevant keywords

Save all metadata in a single youtube-metadata.md file.

Step 7: Organize Files

Structure output in a dated meeting/episode folder:

meetings/YYYY-MM-DD/
├── transcription.txt          # Full text
├── transcription.json         # Timestamped segments
├── 01-clip-name.mp4           # Topic clips (16:9)
├── 01-thumbnail.png           # Landscape thumbnail (1280x720)
├── 01-short.mp4               # Vertical short (9:16)
├── 01-short-thumb.png         # Vertical thumbnail (1080x1920)
├── ...
├── youtube-metadata.md         # Titles, descriptions, tags
└── cuts.txt                   # Cut plan documentation

Step 8: Upload to YouTube

Use Claude in Chrome (MCP browser extension) to automate YouTube Studio uploads:

1. Navigate to https://studio.youtube.com
2. Click Create → Upload videos
3. User must manually select the file (browser security blocks programmatic file uploads)
4. While video uploads, fill in metadata:
   • Title: Use the hook-driven title from youtube-metadata.md
   • Description: Full description with timestamps, links, hashtags
   • Tags: Comma-separated keyword list
   • Thumbnail: User must manually upload (same browser restriction)
   • Audience: "No, it's not made for kids"
5. Click through: Video elements → Initial check → Visibility
6. Set visibility (Public/Unlisted/Private) — always confirm with user before publishing
7. Click Publish

Important limitations:

• Browser extensions cannot upload files to <input type="file"> elements — user must do this manually
• Always confirm before clicking irreversible buttons (Publish, Send, Post)
• YouTube processing takes several minutes after upload; SD first, then HD

Step 9: Promote

After publishing:

• Cut YouTube Shorts (60-90s vertical clips) as teaser content
• Share in relevant communities (Reddit, Telegram, X/Twitter)
• Consider creating a playlist for the series (e.g., "AI Study Group")
• Post short clips first — they drive discovery to the long-form videos

Tips & Lessons Learned

• Split long recordings into topic-based clips — 48 min is too long for YouTube retention. 5-15 min per topic performs better.
• Shorts drive discovery — the counting demo and "elephant problem" are perfect 60s hooks.
• Thumbnails matter most — big text, 2-3 colors, readable at small sizes. Use brand-poster skill for consistency.
• Remove dead time aggressively — login screens, tech issues, coffee breaks, link searching. Keep only content.
• Stream copy (-c copy) for speed when cutting — only re-encode for crops (Shorts).
• Parallel processing — run independent ffmpeg cuts simultaneously.
• Description SEO — include timestamps (chapters), links, hashtags at the bottom.
• Tags — 8-12 keywords mixing broad ("AI") and specific ("ChatGPT count to 200").
• Verify before publishing — always ask user to confirm visibility setting before clicking Publish.