Case study · 04
Cutroom
Turn a codebase into product videos, on open models.
- Role
- Creator
- Period
- Oct 2026 – present
- Stack
- TypeScript, React, Remotion, Gemma, ElevenLabs, Kokoro, Node.js
- Status
- Hackathon
01 The problem
Teams ship every week, but launch videos take days, so most releases go out without one.
Generated videos have a worse problem than effort. They make claims the product does not support.
Cutroom writes the video from the code itself and checks every claim against that code before anything renders.
02 What I built
- cutroom
cutroom: npx cutroom demo
- read repo
read repo: Secrets redacted first
- product brief
product brief: Every feature cites file and line
- storyboard
storyboard: Scenes and narration
- claim audit
claim audit: UI labels, code, numbers, commands, then a Gemma judge
- human approval
human approval: Nothing renders unapproved
- render
render: Remotion, −16 LUFS
- local mode
local mode: Gemma + Kokoro, privacy and egress guards in code
Cutroom starts by reading the repository, redacting secrets before anything else sees the code.
It writes a product brief in which every feature cites file-and-line evidence, then a storyboard. Before rendering, it audits every claim against the code: UI labels, code excerpts, numbers and commands, and finally a Gemma judge.
A person reviews and adjusts the cut in the storyboard and timeline editor, built in React. Then Remotion renders the video, narrated with ElevenLabs, and the audio is normalised to −16 LUFS.
In local mode the whole pipeline runs offline, with Gemma for reasoning and Kokoro for the voice. A privacy guard and an egress guard are enforced in code, not in documentation.
03 Decisions and trade-offs
Decision 01
Chose Auditing every claim against the code over trusting the model's script.
Why: A launch video that shows a feature the product lacks does more harm than no video.
Cost: Extra pipeline stages and slower runs.
Decision 02
Chose A fully offline local mode with guards enforced in code over a privacy promise in the README.
Why: Teams can point Cutroom at private repositories knowing nothing leaves the machine.
Cost: Open models that run locally are smaller than hosted ones.
04 Results
- cutroom · npm downloads / month
- 316
- Source: npm downloads API, 2026-09-27 to 2026-10-26
What shipped
- Built for the DEV Hacktoberfest 2026 open-source AI challenge.
- Published on npm as cutroom (npx cutroom demo), with deploy buttons for Render and DigitalOcean.
- A storyboard and timeline editor for product launch, tutorial and explainer videos.
- Local mode runs fully offline with Gemma reasoning and Kokoro voice.
05 What I would do next
- Launch cutroom.tech.
Want this kind of work on your team?