FOR PAGEMOTOR FORUM MEMBERS
Launch videos from your own website, made by Claude Code
A Claude Code plugin that turns one page of your site into a 30 to 60 second launch film with music, sound and a voice. The graphics are drawn in code, so every word and every screen on it is exactly right, in your own colours and fonts.
launch-video plugin, version 1.3.0
One zip, 51 KB. Install it with the prompt in lesson 3. Checksum (sha256): 4d70ffa488e831c3a1231e051ed11fd6958c3d1278d8a93ecf7453ec7fda75e9
Using OpenAI Codex instead? Codex edition, version 1.3.0
The same skill for Codex, 46 KB. Install steps are at the end of lesson 3; it made two of the films on the comparison page. Checksum (sha256): 739ae87a4b0358ac649c69484c9bd2a8a96ee2241a7090280d4b4dc86d4621c4
Lesson 1
What it makes, and how
You give Claude one page of your website. It reads that page, takes your colours, fonts and images from the site itself, writes a short script using only words that are on the page, and builds a launch film in code. Then it scores it with generated music, places sound effects on the exact frames where things happen, adds a voice, renders it and checks its own work.
What you get
A finished 1920x1080 MP4, 30 to 60 seconds, with the project folder behind it so it can be changed and re-rendered. On request, a 9:16 vertical version for Instagram Reels, TikTok and YouTube Shorts, re-laid out for a tall screen rather than cropped, and a YouTube title, description and thumbnails.
Why it looks right
Every word, button and screen is real HTML, rendered frame by frame, so spelling and interface are exact. No AI video model is used. The only generated pictures are props (a pen, a cup, a tool you use), made as stills, cut out and placed. Claims come word for word from your page.
Already have video? Use it. Hand Claude your own footage, such as a product demo, a screen recording, an older promo or event clips, and it finds every shot, cuts them clean and places them in the film alongside the drawn graphics, taking colour and pace from the footage when your site has little of its own. The words still come only from your page. Use footage you own or have the rights to.
It doesn't invent footage. Nothing is generated of people or places; anything filmed in the result is footage you supplied.
Your footage stays on your computer. The cutting, analysis and rendering all run locally, and your video file is never uploaded. What does leave your machine is only what Claude looks at: small contact sheets of frames from each shot, any single frames it checks, and the shot timings and colours as text. Those go to Anthropic as part of your Claude Code session. ElevenLabs receives only the text prompts for the music, sound effects and voice, never your footage.
What Anthropic keeps. On a Claude Free, Pro or Max plan, a setting decides whether your sessions may be used to train future models. With it off, they are kept for 30 days; with it on, for up to five years. Team, Enterprise and API use are never used for training. Check yours under Privacy Settings before working with a client's confidential footage, and read Anthropic's own explanation for the detail.
Lesson 2
What you need first
The install prompt in lesson 3 checks all of this for you and names anything missing, so treat this as a preview rather than a checklist.
| Need | Why |
|---|---|
| Claude Code (the app or the terminal version), or OpenAI Codex | The film is rendered on your own computer. Claude or ChatGPT on the web cannot do it. |
| An ElevenLabs account with a little credit | Music, sound effects and the voice. Voices are generated with ElevenLabs' Eleven v4 model. Connect it in the Claude app under Settings, then Connectors, or give Claude your own ElevenLabs API key. In Codex, add the ElevenLabs plugin from the ChatGPT plugins directory. |
| Node 18 or newer, ffmpeg, Python 3.9 or newer, git | The rendering and audio tools. If any are missing, Claude Code offers to install them and asks first. |
| A few GB of free disk, not in iCloud or Dropbox | Renders are big while they are being made. Synced folders create duplicate copies mid-render. |
| A strong model | It is a long, careful job: research, script, storyboard, build and several review rounds. For example, Claude Opus 5.5 with Effort set to High or Extra. In the Claude desktop app, pick the model and drag the Effort slider below the message box (it runs Low, Medium, High, Extra, Max, Ultracode). In a terminal, start Claude Code with claude --model claude-opus-5-5 --effort high, or --effort xhigh for Extra. In Codex, GPT-6.1 Sol or GPT-6 Astra at high reasoning effort. |
Referral link: the ElevenLabs links on this page are my referral links. If you sign up and pay through them, ElevenLabs pays me a commission.
Cost: a short film with music and sound effects used well under one dollar of ElevenLabs credit; a longer film with images and a voice used about 3,600 credits. Your Claude plan covers the rest.
Lesson 3
Install it (Claude Code does the work)
- Download the zip from the box at the top of this page.
- Open Claude Code and paste this in:
Please install the launch-video plugin for me from the zip I downloaded. Do every step yourself and only stop when you need me to click or type something. 1. Find launch-video-1.3.0.zip (usually in ~/Downloads) and unzip it into ~/claude-plugins/ so I end up with ~/claude-plugins/launch-video-plugin. 2. Run: claude plugin marketplace add ~/claude-plugins/launch-video-plugin, then: claude plugin install launch-video@elmspark 3. Run the preflight check: bash ~/claude-plugins/launch-video-plugin/plugins/launch-video/skills/launch-video/scripts/preflight.sh, and install anything it says is missing. Ask me before each install. 4. Tell me when it's done, and remind me to restart Claude Code.
Download this prompt as a file
When it says it is done, close Claude Code and open it again so the plugin loads.
Prefer to type the commands yourself?
Unzip the file into ~/claude-plugins/, then in Claude Code run /plugin marketplace add ~/claude-plugins/launch-video-plugin and /plugin install launch-video@elmspark. The preflight check is scripts/preflight.sh inside the plugin folder.
Keep the unzipped folder where it is. Claude Code installs the plugin from that folder, and updates come from it too. Do not unzip it into iCloud or Dropbox.
On OpenAI Codex instead
- Download the Codex edition.
- In Codex (the ChatGPT desktop app or the terminal version), paste this in:
Please install the launch-video skill for me from the zip I downloaded. Do every step yourself and only stop when you need me to click or approve something. 1. Find launch-video-codex-1.3.0.zip (usually in ~/Downloads) and unzip it into ~/.agents/skills/ so I end up with ~/.agents/skills/launch-video. 2. Run the preflight check: bash ~/.agents/skills/launch-video/scripts/preflight.sh, and install anything it says is missing. Ask me before each install. 3. Check you can reach ElevenLabs: search for the tools named elevenlabs_creative_* and list a few voices. If you can't, tell me to add the ElevenLabs plugin from the ChatGPT plugins directory and sign in. 4. Tell me when it's done, and remind me to restart Codex.
Download this prompt as a file
Rendering needs to run outside Codex's sandbox. The renderer changes its own process priority, which Codex's sandbox does not allow. Approve the render commands when Codex asks, or give the session full access in a folder of its own.
Lesson 4
Make your first video
In a new Claude Code session, type something like:
Make a launch video for https://your-site.com/the-page-you-want/
Pick the page that best says what you do; everything in the film comes from it. Claude then works through these stages, and stops to show you the important choices:
- Reads the page and your site's colours, fonts and images.
- Writes the script using only lines from your page.
- Shows three storyboard looks as stills, so you choose a direction before anything moves.
- Builds the film, generates the music, sound effects and voice, and renders it.
- Reviews its own render frame by frame and fixes what is off, for a few rounds.
Your project lives in ~/launch-videos/. Ask for changes in plain English ("slower opening", "a different voice", "end on the booking button") and it re-renders.
Watch it with sound before you share it. The voice is generated with ElevenLabs' Eleven v4 model and defaults to a British English woman; listen for anything that sounds off, and ask for another voice if it does. Claude can check timing and loudness, but only you can judge how it feels.
Lesson 5
Things worth knowing
- It takes a while. A first film is a long session. Let it run and answer its questions.
- It only says what your page says. If a claim is not on the page, it will not appear in the film. Add it to the page first if you want it in.
- It only shows what you actually use. Props stand for your own work, so they come from what your site shows or names. If it is unsure whether you use something, it asks. You can also send a photo of the real thing and it will match it.
- Before and after photos work well. If your site has pairs (a valet, a makeover, a repair), it lines each pair up exactly, so the subject holds still and only the change wipes across.
- Images it generates carry hidden labels naming the AI tool that made them. The plugin includes a script that strips them from a copy if you need a clean one.
- Music rights: the music is generated on your own ElevenLabs account, under your plan's terms. Check your plan covers commercial use before you publish or monetise.
- If something fails, ask Claude to run the preflight check again and tell you what it finds; most problems are a missing tool.
Lesson 6
Updates, licence and credits
Updates: new versions appear on this page. Download the new zip, unzip it over the same folder, then run /plugin marketplace update elmspark and install again. You are on version 1.3.0 if you downloaded today.
Problems and ideas: post in the forum thread where you found this page.
Licence: for PageMotor forum members. Use it and adapt it for your own work and your clients'; the films you make are yours. Please do not pass the plugin on outside the forum. Full terms are in the LICENSE file inside the zip.
Built on the method of launch-video-clone by Pat Simmons (MIT) and rendered with HyperFrames by HeyGen (Apache 2.0). Full credits are in the NOTICE file.