An AI-automated YouTube channel — now agentic.
This is the production pipeline for Tell Me More AI! — a YouTube channel powered by Hermes Agent.
Note: The original hard-coded Python pipeline has been moved to
legacy/. The channel is now fully agentic.
| Script | What it did |
|---|---|
ideagenerator.py |
Searched Wikipedia for podcast topics |
podcastgenerator.py |
Used Selenium to generate NotebookLM podcasts |
podcast2video.py |
Generated slide-show videos + thumbnails |
genai.py / wiki.py / media.py |
AI helpers, Wikipedia scraping, media utils |
yt.py / custom_thumbnail.py |
YouTube upload, custom thumbnails |
Now (hermes-agentic/)
A self-improving AI agent produces 3 Shorts/day:
- Research → AI picks trending topics, checks dedup
- Script → Curiosity-gap format (hook / build-up / reveal)
- Visuals → Pexels videos + images with Ken Burns + crossfade
- Captions → TikTok-style animated word highlighting
- TTS → edge-tts narration
- Upload → Scheduled YouTube Shorts with AI content flags
- Improve → Weekly analytics feed back into the style guide
No human involved — the agent self-improves every week.
See hermes-agentic/ for the Shorts pipeline scripts.
These scripts are designed to be called by Hermes Agent, but can also be run standalone for testing and debugging.
The agent also produces true crime documentary videos (10-15 minute landscape format) using the same Ken Burns, Pexels, and TTS infrastructure as the Shorts pipeline.
See hermes-agentic/true-crime-builder/ for the full skill documentation.
| Script | Purpose |
|---|---|
scripts/build_tc_video.py |
Full pipeline: TTS → Pexels → Ken Burns → Captions → Thumbnail |
scripts/upload_true_crime.py |
Upload with custom description, tags, thumbnail, playlist |
Produces 1920×1080 landscape videos with:
- Ken Burns animated imagery from Pexels (100+ landscape photos)
- Cascaded crossfade transitions (disk-buffered, no OOM)
- Subtitle-bar captions with word-level timing
- Auto-generated thumbnails with title + TRUE CRIME badge
- YouTube playlist integration
The original pipeline that used NotebookLM for podcast generation,
Selenium browser automation, and slide-show video creation lives in
legacy/. It is preserved for reference but no longer
active in production.