Five named structural presets (bold, calm, editorial, energetic, minimal) that set the pacing/ground/exit knobs build.py already reads with a default (brand mode/radius, film gap, music bpm/drop scale, per-beat-type ground/exit/hold) without ever touching brand identity (colors, fonts, logo, name, tagline, domain, url, description stay exactly what brand-from-url.mjs extracted). tools/apply-preset.mjs applies a preset onto brand/brand.json + film/film.json: a beat's own explicit field always wins unless --force, writes timestamped backups before touching either file, and stamps a `_preset` block into both so a later diff explains why gap/mode/radius changed. --list and --dry-run verified against the repo's own Acme example (bold/minimal dry-run output confirmed no writes and correct skip-when-hand-set behavior). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Launch Video Kit
Turn your website into an Apple-style launch video in your own brand, made by your coding agent.
This is the system I use to make our launch films, rebuilt so it works for any product. Point it at your website. It pulls your colours, fonts, logo and copy into a brand kit. Then Claude Code or Codex writes the story from your own words and renders a 30 to 60 second launch film: a hook, your name on the music's drop, hero words cut on the voice, your product at close range, your logo in near silence.
The first version is free and needs no accounts: a placeholder narrator, a synthesised music bed and a synthesised sound palette. Swap in a real voice, music and footage when the story works.
Use this template · Download ZIP · All 11 skills
The example film for Acme, a fictional brand, extracted from its website by the kit and rendered on the free path.
See your own brand extracted with zero install: paste your URL at
launch-video-kit-preview.subasiarhan3.workers.dev
(a hosted Cloudflare Worker, worker/) and get your colours, fonts, logo and name back in
under 10s, before you run a single local command.
What you get
| Included | What it does for you |
|---|---|
| Website to brand kit | npm run brand -- https://your-site.com captures your colours by role, your fonts, your logo, your headline copy and screenshots, plus a preview sheet to check at a glance |
| The film template | A hook plus a shared body, drawn in your brand. The story lives in 1 JSON file; every picture cue lands on the narrator's words |
| 11 agent skills | Story, voice, music, sound, mix, review, motion, 3D and competitor-script import, each a short manual your agent follows |
| A free audio path | Scratch voice, forced alignment, a scratch music bed on the bar grid, a synthesised sound palette and a -14 LUFS master |
| Paid upgrades, on your account | Narrator voices, music, sound kits and generated footage through Higgsfield, with the cost stated before every job (costs) |
| A review loop | A script that measures a render (cuts, static runs, loudness, beat grid) and writes a filmstrip and contact sheet, plus the eye checklist |
| A worked example | Acme: a fictional website, its extracted brand kit, 3 product screens and the finished film script |
Start here
-
Click Use this template and clone your copy. Keep it private if you add unreleased product screens. No GitHub account handy, or want to skip "Use this template" and the clone entirely?
npx launch-video-kit init my-launchscaffolds the same starting point from one command. -
See your own brand pulled out before installing anything else:
npm install # this package's only dependency: puppeteer npx launch-video-kit brand https://your-site.com --openWrites
brand/preview.html: your colours, fonts, logo and headline copy, at a glance. No ffmpeg, no uv, no whisper-cpp, no torch - just Node. (npm run demo -- https://your-site.comis the same command, kept for muscle memory.) -
Install the rest of the toolchain to render the actual film (setup, about 10 minutes on a Mac):
brew install node ffmpeg uv whisper-cpp uv venv .venv && uv pip install -r requirements.txt npx launch-video-kit doctor # confirms every tool above is really on PATH before you render -
Open the folder in Claude Code or Codex and paste:
Read AGENTS.md and README.md. Make a launch video for my product. Start with the launch-brand skill on https://YOUR-SITE.com and show me brand/preview.png; fix anything that reads wrong. Then use launch-film: draft the script from my website's own words, show me the beats and a contact sheet before rendering, and render the free version first.
Bring 3 or 4 screenshots or short screen recordings of your product. They carry the film.
Or run the example by hand
npm run brand -- https://your-site.com # optional: your brand instead of Acme's
.venv/bin/python skills/launch-voice/scripts/scratch_voice.py film/film.json
.venv/bin/python skills/launch-voice/scripts/prep_lines.py film/film.json --align force # precise word times; uv fetches torch once
.venv/bin/python skills/launch-score/scripts/scratch_bed.py film/score/bed.wav --bpm 120 --intro-bars 3
film/render.sh type-only
The film lands in film/renders/type-only.mp4. With your brand extracted, it is Acme's script and screens in your colours, fonts and logo. Replace the lines in film/film.json and the screens in film/assets/screens/ to make it yours.
How a film is built
website ──> brand/brand.json ──┐
├──> film/build.py ──> HyperFrames render ──> film/audio.py ──> master.mp4
film/film.json (lines, beats) ─┤ (every cue is a spoken word's time) (music on the cut, voice,
voice takes ──> word times ────┘ sound, -14 LUFS)
The hook is anything before the cut: 2 lines of type (free), your own footage, or generated footage. On the music's drop it cuts to the body, which is built from 6 beat types:
| Beat | On screen |
|---|---|
name |
Your logo and name punch in from 4x on the drop; the tagline writes in word by word |
hero |
Full-frame words on a dark, accent or light ground, each snapping in on its spoken word, leaving by a dive, a collapse or a cut |
product |
Your real screen in a window; the camera pushes into the part the voice names, with an optional cursor click and a chip |
blocks |
Your features land on their names, then close into 1 |
switch |
A card that turns on: "Feedback, on autopilot." |
end |
The lockup, then a quiet hold |
The body is shared by every hook, so testing 5 hooks is 5 small JSON files, and a change to the body reaches every film.
Rules we learned the hard way
- Type that lands 0.1 to 0.3 s after its word reads as off beat. Time it by forced alignment, never by guesswork.
- 1 narrator for the whole film. A second voice for a new line breaks it, even when it is clearer.
- Show frames before building. A full render on the wrong look wastes a day.
- The whole app window at thumbnail size says nothing. Push into the one part the voice names.
- No text card where the animation already says it, and no monotone stretch longer than a few seconds.
- When photoreal people look fake, go stylised instead of cheaper.
More in lessons.
Cost, compared
No subscription, no seat fee, no credit-card wall to start. Named prices, not "cheaper":
| Tool | Entry price | What it buys |
|---|---|---|
| launch-video-kit, free path | $0 | Brand kit, type-only hook, full body, scratch voice/music/sound, mixed and mastered - no account |
| launch-video-kit, paid upgrade | ~$1.50-3 for a typical first film (30-60 Higgsfield credits, costs) | Real narrator voice, real music take, a generated sound kit - on your own account, cost stated before every job |
| Waymark | $350-399/month (Starter, annual or monthly) | 2-4 final downloads/month, broadcast stock footage, ad-buy integration |
| InVideo AI | Free tier watermarked (10 min/week); paid ~$17-28 to ~$900-1,000/month | Credit system burning 3,000-30,000 credits/project |
The full, sortable comparison (these two plus shipvideo, Arcade, Tella and lemomo-ai/lemo-opuscar) is live at launch-video-kit-compare.subasiarhan3.workers.dev/compare (site/).
The CLI
npx launch-video-kit <command> wraps the scripts above so a fresh clone doesn't need to learn the folder layout first:
| Command | Does |
|---|---|
init <dir> |
Scaffolds a working copy into <dir> (shallow clone, history reset) without visiting GitHub or clicking "Use this template" |
brand <url> (alias demo) |
Runs the brand extractor; --open launches preview.html when it is done |
render [hook...] [--format landscape|vertical|square|all] |
film/render.sh under the hood - builds, renders and mixes; --format all writes the 16:9, 9:16 and 1:1 cuts side by side |
doctor |
Checks ffmpeg, python3/.venv, uv and whisper-cli and prints the exact per-OS install command for anything missing, instead of a render failing halfway through |
launch-video-kit --help or any subcommand with --help prints its own usage.
Three more scripts under tools/ cover the rest of a real workflow, each with its own --help:
tools/capture-screens.mjs <url> --routes routes.jsondrives headless Chrome to each listed route, waits out loading skeletons, and writes croppedhero.png/full.pngscreenshots straight intofilm/assets/screens/- the 1 manual step ("bring 3-4 screenshots") this kit cannot otherwise do for you.tools/batch.mjs/film/render-all.sh <manifest.json>renders a whole portfolio in one pass: for every{url, screenshots, lines}entry in a manifest, it runsbrand-from-url.mjs→build.py→render.sh, writingrenders/<product>/<hook>.mp4per product and arenders/batch-report.jsonsummary.tools/import-script.mjs <competitor-export.json>converts a flat, single-scene script exported from a shipvideo/Waymark/InVideo-style tool into this kit's 6-beatfilm.jsongrammar, ready to render on the free path immediately. Seeskills/launch-import/SKILL.md.tools/localize.mjs <lang1,lang2,...> [--hooks name...]translatesfilm/film.jsonand everyhook.jsoninto one or more languages (free, no API key - the unofficial Google Translate endpoint, cached by text hash) and writesfilm/i18n/<lang>/, re-anchoring every on-screen word-timing key so the translated film still cuts in the right place.tools/render-lang.mjs <lang> [hook...] [--format ...](orfilm/render-lang.sh) then stages that language over the livefilm.json/hooks/, renders through the real pipeline, writesrenders/i18n/<lang>/<hook>*.mp4, and always restores the English originals.
What is in the repo
skills/ 11 skills: launch-film (start here), launch-brand, launch-import, story, voice, score, sound, mix, review, world, motion
bin/ the npx launch-video-kit CLI (init, brand/demo, render, doctor)
tools/ brand-from-url.mjs (brand extractor), capture-screens.mjs, batch.mjs, import-script.mjs, localize.mjs, render-lang.mjs
brand/ your brand kit (ships with the Acme example)
film/ the template: film.json, hooks/, assets/screens/, templates/body.html, build.py, audio.py, render.sh, render-all.sh
examples/ the fictional Acme website the example brand was extracted from
docs/ setup, costs, rights
Use it for your own brand. The extractor is meant for your own website, and the rights notes cover people, footage, music, fonts and claims.
Built by Tim Scheuer. License · Third-party notices
