LOCAL SETUP / REVIEW BUILD

Your voice.
On your computer.

A coding-agent skill with a real microphone studio and local voice models.

Start with your agent

Extract the archive. Add its narrate-video folder to your agent’s local skills location, following that agent’s installation instructions. The folder contains SKILL.md, scripts, references and local verification checks.

Then ask your agent to use $narrate-video and open the voice studio. The agent should read the bundled local setup guide, reuse a compatible environment or prepare an isolated one, and start the loopback server.

The browser flow

Choose your narration languages.
Review compatible cloning models and the recommendation.
Download and verify the selected model.
Record a short passage, or reuse a saved recording.
Generate and listen to auditions in every selected language before saving the profile.

What you need

The bundled speech adapters currently require Apple Silicon macOS, Python 3.12 with MLX-Audio 0.5.1, and FFmpeg. Core video rendering also uses Pillow. Model weights are separate downloads; no cloud API key is required. Other platforms can use supplied audio with the renderer.

What remains yours

Recordings, transcripts, model weights and generated voice files stay in the private data folder selected by your agent. They are not bundled into the public skill. The website does not upload microphone audio or connect to a private voice service.

For developers

Run the bundled scripts/verify.sh locally before changing or redistributing the skill. Use the full bundled documentation for narration identity, language eligibility, source attribution and media review.