Content Pipeline

Turns a raw .mov screen recording into a full content package: a narrated YouTube video with a branded intro, animated concept cut scenes, an SRT caption file, a thumbnail, a blog article, and YouTube metadata.

It runs as a Claude Code skill (shipped in .claude/skills/content-pipeline/) backed by an idempotent, multi-stage Python pipeline.

What it produces

Output	Stage	Description
content-ideas.md	I	Ranked trending-AI-tool ideas to record next
frames/	1	Deduplicated screenshots from the recording
captions.json	2	Claude Vision narration per frame
output.mp4	3+4	Composed video with AI narration and music
thumbnail.png	5	1280x720 YouTube thumbnail
youtube.md	6	Title, description, and tags
captions.srt	7	SRT subtitles
article.md	8	Full tutorial blog post
output_with_intro.mp4	0	Final video with branded intro

Pipeline stages

Run in order (each stage is idempotent, so you can re-run safely):

1. ideate.py — pick what to record based on trending signals
2. 01_extract_frames.py — extract and dedupe frames
3. 02_caption_frames.py — Claude Vision writes first-person narration per frame
4. 03_generate_audio.py — ElevenLabs TTS + background music
5. 04_compose_video.py — cut and compose the video
6. 05_create_thumbnail.py — build the YouTube thumbnail
7. 06_youtube_metadata.py — generate title, description, tags
8. 07_generate_srt.py — transcribe audio to SRT
9. 08_generate_article.py — write the blog post
10. 00_create_intro.py — prepend the branded intro (run last)

See .claude/skills/content-pipeline/SKILL.md and the reference/ docs for full details.

Setup

pip install -r requirements.txt
cp .env.example .env   # then fill in your keys

Required keys (see .env.example):

• ELEVENLABS_API_KEY and ELEVENLABS_VOICE_ID — text-to-speech and transcription
• AI_GATEWAY_API_KEY — Claude access for captioning, ideation, and writing

Notes

• Media files (.mov, .mp4, generated frames/, audio/, etc.) are gitignored — they are large and regenerated per run.
• Never commit your real .env; it holds API keys.