ContentGenie

Provenance: ContentGenie is downstream engineering based on ShortGPT by Ray Ventura. The upstream MIT copyright is preserved. See ATTRIBUTION.md for the contribution boundary used by this portfolio.

ContentGenie is a production-focused workspace for creating original YouTube Shorts with researched scripts, editorial review, voiceover, word-timed captions, generated visuals, sound design, licensed gameplay automation, and encoded-file quality checks.

Run Locally

python runContentGenie.py

The Next.js studio runs at:

http://localhost:31415

The launcher starts the private FastAPI service automatically and rebuilds the frontend when its source changes. Node.js 20 or newer is required.

For frontend-only development:

cd frontend
npm install
npm run dev

Main Features

• Schema-constrained story packages with Gemini.
• Grounded research for facts and history content through Gemini Search, with Wikimedia and date-specific "On this day" sources as a resilient fallback. If both are unavailable, factual generation stops instead of publishing unverified claims.
• Production mode with a second editorial pass that scores the hook, retention, originality, clarity, payoff, and advertiser safety.
• Similarity checks against recent channel output to reduce repetitive, mass-produced results.
• Creative direction controls for the episode brief, audience, tone, editorial angle, and target duration.
• EdgeTTS or Gemini Flash TTS voiceover.
• Whisper-timed captions, including traditional and color-bounce styles.
• Generated Z-Image overlays with configurable size, screen position, pacing, and subtle motion.
• Timed sound effects from the local assets/sfx library.
• Automatic licensed-gameplay discovery from Wikimedia Commons, YouTube Creative Commons, and Internet Archive, with a key-free YouTube fallback.
• Strict CC0/public-domain/CC BY commercial-use gates, publisher-policy checks, source checksums, saved license evidence, and automatic description attribution.
• Motion, clarity, exposure, and vertical-crop analysis that assembles several high-retention gameplay moments, snapping cuts to caption pauses and SFX beats instead of selecting one random continuous clip.
• Mixed, parkour, racing, satisfying, and action footage styles with balanced/high/extreme pacing and recent-segment reuse prevention.
• Manual background video, background music, captions, image layout, and normalized audio mix remain available.
• Post-render checks for vertical aspect ratio, resolution, duration, audio, frame rate, and file integrity.
• A JSON production manifest next to every video with the script, research sources, prompts, quality reports, media-rights confirmation, and human review checklist.

Production workflow

1. Add commercially licensed music in Media Library. Use Discover free licensed gameplay to inspect or pre-download automatic footage if desired.
2. Add your Gemini key in Settings. A YouTube Data API key is optional because the slower key-free CC metadata search remains available.
3. In Create, leave Automatic licensed gameplay enabled, choose a footage style and cut intensity, and provide the episode's creative direction.
4. Review the media-rights acknowledgement and create the Short. The automatic library replenishes itself when eligible footage is low.
5. Watch the entire export and inspect its .json manifest and generated gameplay credits before uploading.

The tool does not promise views, monetization, or income. You are responsible for factual accuracy, originality, advertiser suitability, copyright, synthetic-content disclosure, and compliance with YouTube policies.

Output files

Each successful render creates three files in videos/:

• .mp4 — the finished vertical Short.
• .txt — upload title and description.
• .json — production manifest and review checklist.

Tests

python -m unittest discover -s tests -v

Requirements

Install dependencies from requirements.txt, configure API keys in ContentGenie's Settings tab, and make sure FFmpeg is available.