Case study · Personal project
Teacher: explainer videos from a script
A personal tool that turns a JSON script into an explainer video with 3D scenes and a neural voice, timing each object to the exact word the narrator says.

- 1 script→ 1 video
- 5lessons produced
- 0API keys
The problem
I wanted to put together a personal course on AI and automation in business finance processes, with a short video for each lesson. Recording and editing every video by hand takes a lot of time, and every change to the content means editing again.
The approach
Describe each video as data. A JSON script defines the narration, the scenes, and which object appears on which word; the code generates the rest.
How it works
- The voice comes from Microsoft Edge TTS: neural voices, free, with no API key.
- The 3D scenes are built with Three.js.
- Word-level sync: each object appears exactly when the narrator says its word.
- A headless browser renders the frames and ffmpeg combines them with the audio into the final video.
- Everything is generated by code, so the same script produces the same video on any Windows PC.
Decisions
- Data over editing: changing a sentence means changing the script and rendering again.
- Free, local tools, with no accounts or keys.
- I tried a 3D presenter with lip sync and dropped it: it pulled attention away from what was being explained.
Outcome
With Teacher I produced a five-lesson personal course on AI and automation in business finance processes such as month-end close and bank reconciliation: what is worth automating, what is better assisted, and what should stay with a person.
Each lesson combines a quiz, a free-recall exercise, an interactive animation and a video made with the tool.