Open source · Claude Code
inkwell
A product URL in, ad videos in every aspect ratio out. Or a narrated launch film.
It reads a product page for the imagery and brand colours that page already publishes, then renders the same ad at every size a platform accepts. Each ratio is laid out separately rather than letterboxed from one master, because a vertical ad with bars down the top and bottom throws away the part of the frame that does the work. For a launch, it makes a narrated film instead: one voice, a scored take per line, and sound design under every beat.
What it will do
- Checks for Node 20+ and ffmpeg, and asks before installing either
- Clones inkwell and installs its agent skill
- Renders one test ad from a product URL you give it
5 to 10 minutes, mostly the first npm install. It asks before anything system-wide, and never asks for a password or API key.
Set up inkwell for me (https://github.com/debashisn94/inkwell), an open-source tool that turns a product URL into ad videos in every aspect ratio, built on Remotion. 1. Check for Node.js 20 or newer, ffmpeg and ffprobe. If any is missing, tell me what and ask before installing it. 2. Clone https://github.com/debashisn94/inkwell into ~/inkwell (or a folder I name), then read README.md and docs/getting-started.md. 3. Install the skill so you can use inkwell later: if you are Claude Code, copy skills/inkwell to ~/.claude/skills/inkwell. Otherwise run `npx skills add https://github.com/debashisn94/inkwell --skill inkwell`. 4. Ask me for a product URL, mine or one I like. From the inkwell folder run `node tools/new-ad.mjs <url> --out=ad`, then `cd ad && npm install && node ../tools/render-ad.mjs Square` to prove it works. The first render downloads a headless Chrome of about 90 MB. 5. Show me ad/ad.json and explain that every line marked REWRITE is a starter. The copy is mine to write, and nothing in it should be a claim the product's own site does not make. Skip the narrated-film format (it needs Python and VoxCPM) unless I ask for it. Ground rules: only download from the GitHub links above. Ask me before using sudo, installing anything system-wide (Homebrew, Node, Python, ffmpeg) or changing files outside the install folder. Never ask me to paste a password or API key into this chat. When you finish, tell me what you installed, where it lives, and the one command I run to use it. If you cannot run terminal commands, give me the exact steps to run myself instead.
git clone https://github.com/debashisn94/inkwell.git MIT licensed. Needs Node and ffmpeg, plus Python and VoxCPM for narration. No API keys, no accounts.
A 30-second ad for Holt, rendered from one definition. Every line and colour came from the product’s own page.
01Formats
One definition, four cuts.
Type scales off the short edge, so a headline is not enormous at 16:9 and unreadable at 9:16. Safe areas match what the platform actually covers. The stack direction flips between portrait and landscape, so neither is a crop of the other.
- Reel9:161080 x 1920 Reels, Shorts, TikTok, Stories
- Feed4:51080 x 1350 Instagram and Facebook feed
- Square1:11080 x 1080 Feed, LinkedIn
- Wide16:91920 x 1080 YouTube, LinkedIn, X, pre-roll
02How it works
Three commands, and the one that matters is the middle one.
# 01 Research the product$ node tools/new-ad.mjs https://your-product.com --out=ad# 02 Write the ad$ edit ad/ad.json# 03 Render every size$ cd ad && npm install && node ../tools/render-ad.mjs - 01
Research the product
Reads the page and writes brand.json: name, positioning, headings, the imagery it publishes, and a palette sampled from that imagery. Then drafts an ad.json with every line marked for rewriting.
- 02
Write the ad
Nine scene types. Four carry the copy: hook, feature, proof, call to action. Five change inside themselves: a prompt that types and gets its answer, notifications stacking, a chart that lands on its number, a marquee of the models or integrations the site names, and isometric line figures. The copy is the part that decides whether the ad works, and it is the part the tool deliberately does not finish for you.
- 03
Render every size
Bundles once and renders all four formats. Adding a fifth is one entry in a list, because the layout engine derives type scale, safe areas and stack direction from the dimensions.
03The film
New in v0.3
Or a narrated launch film.
An ad has fifteen seconds. A launch needs two minutes: the problem shown happening, why the obvious fix fails, the reveal, how it works, and the offer. Same research step, a different output, voiced and sound-designed.
- 12scene types, from a live terminal to a lock slam and an architecture flow
- 1narrator for every line, so the voice never changes mid-film
- 15synthesized sounds, each scene firing its own on its own beats
- 01
Write the film
Scaffolds the project, pulls the palette from the product’s own site, and lays out a structure: problem, the obvious fix and why it fails, reveal, how it works, offer, action. Each scene carries its on-screen copy and one line of narration.
- 02
Voice it
Designs one narrator, then clones every line from it with its own delivery note. Each line gets two or more takes, each take is transcribed back and checked against the narrator’s voice, and the most expressive take that passes is kept.
- 03
Render picture and sound
One pass renders the picture, the narration and the sound design, then masters it to -16 LUFS. The stems come out separately, so music goes in under the voice without re-rendering anything.
# 01 Write the film$ node tools/new-film.mjs https://your-product.com --out=film# 02 Voice it$ python ../tools/film-voice.py# 03 Render picture and sound$ node ../tools/render-film.mjs --stems Scene types, the voice scoring, and mixing music under the stems are in the film guide.
04Decisions
Eight decisions that took more than one attempt.
Each of these was wrong first, in a way that looked plausible until it was measured.
- 01
It reads the page head, not the images
Most product sites render their imagery in JavaScript or as CSS backgrounds, so scraping image tags returns nothing on exactly the polished sites worth advertising. The og:image is reliably present, correctly sized, and chosen by the owner to represent the product.
- 02
The palette comes only from the brand’s own assets
Sampling every image on a page sounds more thorough and is wrong. On a portfolio or a customer-logo strip those are other companies’ marks. Tested against a site carrying client logos, the extracted accent came back as the client’s teal instead of the brand’s amber.
- 03
Colour is read at high resolution or not at all
Sampling a 1200 by 630 banner on a 32 by 32 grid averages each cell across roughly 37 pixels of source, which is wider than the lettering in most logos. The accent is destroyed before any of the colour logic runs. It samples at 260 and the extracted colour lands within a couple of percent of the real one.
- 04
One narrator, not eleven
Designing each line from a written voice description sounds like the natural way to direct a narrator. Tested on the first film, every line came back as a different person. The narrator is now designed once and saved, and every line is cloned from that recording. Delivery changes per line. The voice does not.
- 05
The narration sets the edit
Scenes were first timed by hand, and every re-voiced line broke the timing of everything after it. Now each scene lasts as long as its animation needs or its line needs plus a breath, whichever is longer. Re-voice a line and the film retimes itself.
- 06
Assets keep their own shape
An og:image is a 1.91:1 banner, usually with the product name set across it. Fitting that into a portrait box with a cover crop takes the wordmark off both ends, so every asset is measured and given a box that matches it.
- 07
Pacing comes from a real ad, and only from its cuts
Measured against a professional 30-second ad, which changes picture every 1.9 seconds, the template changed every 4.6. study.mjs finds the cuts in an ad you admire and gives your scenes its proportions. The first version split the ad evenly wherever no cut was nearby, and cut a 7-second end card in half. That long hold is the rhythm being measured, so boundaries now land only on real cuts.
- 08
Line figures redrawn as a function of the frame
The isometric figures use the geometry of Hairline, an MIT library by Lucas Marques. Its figures follow the pointer and step a spring on every animation frame, which cannot render here: Remotion draws frames out of order and in parallel, so anything that depends on the previous frame comes out jittered. Only the pure geometry is vendored, and each figure is redrawn from scratch per frame.
05Limits
What it does not do.
There is a write-up of the part that took three goes, including why the first version measured perfectly and still looked like a slide deck: read the build notes.
- It drafts ad copy from the page and marks every line for rewriting. It can tell you what a product calls itself. It cannot tell you who it beats or what the objection is, and those are what make an ad work.
- Every ad still shares one visual grammar: type on a gradient, with the product’s image or a drawn widget beside it. Nine scene types make it vary inside an ad. Across forty ads it would still look like one house style.
- The interface moments are drawn, not recorded. A typed prompt, a notification, a chart: generic versions in the brand’s colours, not the product’s real screens. A real screen recording or a before and after wipe still has to be made per product.
- Films are 16:9 only. Their diagrams and side-by-side layouts need a wide canvas, and a vertical cut of a film is a different edit, not a reflow.
- Narration runs locally on VoxCPM, which needs a capable GPU or Apple silicon and Python. The takes are scored automatically, but the score cannot hear a flat read. Someone still has to listen.
- Remotion is free for individuals and teams of three or fewer, and needs a company licence above that.