Explainroo is an open-source project that lets a coding agent turn a topic into a narrated explainer video on your own computer
Explainroo is a MIT-licensed project by Vincent Scheib that lets Claude Code, Codex or another coding agent turn a topic into a narrated explainer video on the user's machine, with an open voice model, open timing model and a canvas renderer that need no cloud account.
Image: GitHub
Why it mattersA short explainer video that an agent can draft without a paid cloud rendering service is the piece most AI-video tools charge monthly for, so a local pipeline the agent drives itself gives a small team a way to make product demos and topic videos at the cost of its existing coding agent plan.
A small team that wants a short explainer video for a product usually ends up paying a monthly rate to one of a handful of cloud video tools, because the local alternatives are either hard to drive or cannot talk. Explainroo, a project released on 25 September 2026 by Vincent Scheib under the GitHub user vincentsch, takes a different route: it ships a pipeline a coding agent can read and operate on its own, so Claude Code, Codex or Pi can turn a topic into an MP4 on the user's own machine. The repository had 403 stars and 12 forks ten days after launch, and the licence is MIT.
The usual way in is to clone the repository and paste a short prompt into the agent, naming the topic. The agent reads the project's AGENTS.md, sets the project up, writes a script and a scene file for the video, and runs the checks the project ships. The output is one MP4 file and one self-contained HTML player.
What runs on the local machine
Three open models and tools do the work. Kokoro-82M, an open text-to-speech model, reads the script with one of 28 voices. Whisper, OpenAI's open-source speech recognition model, listens to the recording and records the time of each spoken word. A headless Chrome draws the frames on a canvas in vanilla JavaScript, and ffmpeg joins everything into the final MP4. The first-time model download is about 400 MB, and the README states that no graphics card is required.
Scheib's README says the pipeline produces vertical films for TikTok, Reels and Shorts, as well as 16:9 for YouTube, 4:5 for Instagram and LinkedIn, and square. Captions light up word by word on tall, 4:5 and square sizes, and the layout keeps text clear of the Reels and TikTok buttons. There are five looks: paper, clean, chalk, blueprint and midnight. A pace setting between 0.7 and 1.6 speeds the voice, the pauses and the animations without rerecording.
The product-demo mode and the stated limits
Explainroo also writes product demos, the short videos software companies make to show their app on their own site. The agent rebuilds the app's screens from its code or its website, with the same colours, fonts and button labels, then a cursor clicks and types through them. The repository ships worked demos for the author's own products Unspar and Vroni, with their source files in the examples/ folder.
Scheib names four honest limits in the README. The agent cannot watch a video, so Explainroo offers it other ways to check: stills of each scene, a sheet of small frames, a layout check that finds cut or overlapping text, and a speech check that finds mispronounced words. A long video spends a large share of the agent's usage plan, because one subagent writes each shot's scene file of 1,000 lines or more and then critic waves review every shot. The project is developed and tested on Linux and the README says Mac and Windows have been tested less. And if the agent asks for AI-generated illustrations through OpenRouter, those image calls are paid for by the user through OpenRouter.
For a founder, marketer or product manager who already pays for Claude Code, Codex or Pi, Explainroo trades a monthly video subscription for the agent's own token budget and a one-time model download. The ten-day traction is real: 403 stars, 12 forks, and the author pushes to the repository daily.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.