Script → voiceover → scenes → MP4
Narrated explainers, agent-produced
A complete local pipeline that turns a script and storyboard into a rendered, narrated MP4 — local text-to-speech and Remotion scenes, fully agent-drivable. No video model, no paid voice, no studio.
The pipeline
The order matters — narration durations drive scene lengths, so the voice is made before the picture.
- Script + storyboard as a mesh node beside the target page: per scene VOICEOVER · ON SCREEN · OVERLAY · GEN PROMPT. Approved before anything is built.
- Voiceover — embedded Python turns
scenes.jsoninto per-scene WAVs plus adurations.jsonvia local piper TTS. That durations file is the contract with the video template. - Scenes as code — one TSX component per scene, animation beats fired at the second the phrase is spoken, captions from the storyboard, palette matched to the target space.
- Render, then embed the approved MP4 into the target page.
What ships
/explainer-videoskill — the end-to-end procedure, including the render gotchas.Pipeline/SynthesizeVoiceover— embedded Python: script → WAVs + durations. Swap in a hosted voice by regenerating the same WAVs.Template/*— a pinned Remotion 4 project: composition, per-sceneSequence+Audiowiring driven bydurations.json, a shared theme, and seven worked-example scenes to copy from.
First production use: the Agentic Engineering Module 1 intro — 2:03, 1080p, 9.9 MB. See the Guide for the full walkthrough.