Turn a topic into a narrated, sourced explainer video

A developer built a Claude Code skill that turns one topic into a sourced script, voiceover and code-rendered 1080p60 film in one chat, and shipped seven sample films made with it.

Done withClaude Code· Oct 28, 2026

What they did

The author, who posts as @Berryxia, packaged a Claude Code skill called vibe-knowledge. Given a single knowledge topic in chat, Claude Opus picks the angle, checks facts so each line of narration has a cited source, writes the script, generates a voiceover through Xiaomi's MiMo voice API, and writes the animation as code that a three.js engine renders frame by frame to 1080p at 60fps with ffmpeg. A QA pass uses OCR to check the on-screen text, and the run ends with a publishing bundle. There are 44 visual styles. An optional tier calls the Renoise MCP server for generated stills and motion shots, which are labelled as AI illustrations in the film. The README says the project itself was orchestrated with Claude Code.

How it went

The repo ships seven sample films made with it, all in Chinese, between 1:10 and 3:12 long, on topics such as how a rocket reaches orbit, what money laundering is and how many folds of paper reach the moon. One episode is rendered twice from the same script, once in the free code-drawn tier and once in the enhanced tier, to compare them. The author gives no figures for time or cost per film.

Worth knowing

Setup is heavy: two skills, a Python venv with Playwright and Chromium, ffmpeg, a MiMo API key, and macOS for the OCR check, which uses Apple's Vision framework. The enhanced tier spends Renoise credits. The default narration voice is a bundled AI voice; cloning your own requires a recording of yourself or one you are licensed to use. The repo was a day old when found.

Try it

With Claude Code· 8 parts to fill in

Make a short explainer video about [topic] for [audience], about [length] long, in [language] and in a [visual style] look. First research the topic and give me a script where every factual line cites a source. Stop and let me approve the script. Then generate the voiceover with [my TTS tool], write the animation as code and render it with [ffmpeg / my renderer] at [1080p60], and check the on-screen text against the script. Report anything you could not verify instead of inventing it.

Square brackets are the parts to fill in. The prompt is ours; the story is theirs.

Where they said it

Related