Builder's Toolkit/Open source · Runs on your machine
Trailers, walkthroughs, help videos. From your live app.
Describe the video. Studio records the real product, writes the script, narrates it in 84 languages, and hands you a 1920×1080 master. Agent Ready for Cursor and Claude Code.
Free · MIT · macOS or Linux
What you can make
Every video your product needs. No video team.
Pick the job. The story structure, the plates, and the length follow. Your product plays the lead.
The showcase cut
This is
Walkthrough Studio.
The desk we use to ship every walkthrough, brief to master
Story the agent drafts for a trailer
- 01What changed
- 02The new thing, live
- 03Proof
- 04Try it today
Agent Ready
Tell Cursor or Claude. It makes the video.
The agent discovers your brand from the live site, drafts the script and the clicks, re-records after a redesign, and recuts for a new channel. You approve. Nothing records or renders without you.
- Discover brand: logo, five colors, type, motion, from your URL
- Draft script, flow, and narration from a one-paragraph brief
- Re-record when the product changes. Voice is reused, not re-bought
- Branch focused feature videos from one overview
You say
Cursor does
- ○discover-brand from app.acme.com: lockup, 5 colors, type
- ○analyze: 31 controls mapped on /billing
- ○drafted 5 chapters, 2 proof plates, 11 lines
- ○waiting for script approval
Your video project
- video.config.jsonScript and plateskept
- flow.mjsWhat gets clickedkept
- narration.mjsWhat gets saidkept
- brands/acme/Logo, colors, motionkept
- acme-launch.mp4The masterkept
Plain files in your repo. Every change is a diff you can read, revert, and commit.
Watch it work
Your app records itself. You direct.
The real desk, playing the real cut. Brand plates render live. Scrub the timeline, switch the tempo.
English · en
Invite a teammate in under a minute.
84 languages
Record once. Speak to every market.
Choose a language in the brief. Captions, plate copy, and narration follow it. Numbers and must-say lines stay exactly as you wrote them. Your recording is reused, so the second language costs minutes, not a reshoot.
IIElevenLabsEleven v4 · narration and music bedYour key, your account
You stay in control
Describe it. Approve it. Then it records.
Five short steps. The agent drafts from them. Nothing records until you approve the script, and nothing ships until you approve the cut.
New video
The agent drafts the script from this brief. Nothing is recorded until you approve it.
Basics
What to show, for whom, and about how long.
Description *
Start URL *
Video language *
Your brand, not a template
Looks like you made it. Because you did.
Your logo, your colors, 38 live backdrops. The agent pulls the brand from your site. You pick the mood.
Atmosphere · Aurora
Every walkthrough.
One studio.
Slow ribbons of brand light. The film opens on it.
Everywhere you post
One recording. YouTube, LinkedIn, Reels, docs.
The wide master, a square cut, a vertical cut, a light 720p for email, and a GIF for the README. Captions and chapters included.
The master. Stakeholder reviews, the website, YouTube.
Start in minutes
Browser, terminal, or your agent.
One command checks Node, npm, and ffmpeg, then opens the desk. Prefer the terminal? Four commands. Prefer your agent? Ship the skill with your repo.
Before you clone.
No. You describe the video in a five-step brief. The agent drafts the script, Studio records your live app, adds narration and music, and renders the master. A timeline is there when you want a precise retime, and a plain-language note does most edits.
Trailers for a launch, feature walkthroughs for the changelog or a sales follow-up, help and onboarding videos for your help center, and sales demos. Each type has its own story spine, so a help video does not open like a trailer.
Studio shows an Agent Ready switch when either CLI is signed in. The agent discovers your brand from the live site, drafts the script, flow, and narration, applies your notes with the smallest safe scope, and branches feature videos from an overview. Drop SKILL.md into your repo and the agent can run the whole pipeline without opening the browser.
Pick the language in the brief. Plate copy, captions, and narration are written in that language, numbers and must-say lines stay verbatim, and ElevenLabs v4 speaks it. The recording of your product is reused across languages.
Nowhere. Capture, renders, and the master stay on your machine. Demo logins live in kit/.env, never in the project file. Narration and the music bed call ElevenLabs only when you add a key. Studio opens without one.
Studio is MIT licensed and free. ElevenLabs is pay as you go on your own key: a 3-minute video is about 1,300 characters of speech and 3 minutes of music. macOS or Linux, Node 20 or later, and ffmpeg.