Preview gallery
Similar Skills
View allWhat you get
- A 16:9 WebP with your mark at the heart of a near-white line field, every letter of brand and headline exact
- The exact prompt it sent plus the saved FAL task file; rerunning resumes that request rather than buying a new one
- A review of spelling, the 6% safe margin and the 16:9 frame before the card reaches you
Good fit
- APIs, search products and other developer tools
- Docs pages and release posts that should stay quiet
- Brands built around one strong blue
Not for
- Photos, crowds or busy scenes
- Numbers, charts or table data
- Rebuilding a real company's logo or campaign
About this Skill

The look is restraint. The background is almost white; thin, pale curved lines bend in from the sides and gather low across the frame like a horizon. In the open space they leave sits a centered lockup, your logo and wordmark in cobalt blue, with your short headline. If you have no mark, the agent draws a simple one, writes the prompt, sends it to GPT Image 2 on FAL and reads the result letter by letter before handing over the WebP.
How it works
- Agree on the name, one headline and one picture idea; drop anything that blurs at thumbnail size
- Lay out the card from references/visual-system.md and fit your colors to its four style anchors
- Strip out outside names, marks, faces, screens and slogans, then invent a simple center mark for your brand
- Write a prompt about visible things only: frame, layout, light, palette, subject and type position
- Send one request with the bundled FAL helper and keep the task file it writes
- Read each letter, check the 6% margins and the 16:9 frame, and only then run validate-output.mjs
- If a flaw remains, propose one prompt change and wait for your OK before a second paid request
Checked before delivery
- validate-output.mjs passes with --require-format webp --require-16-9, and the prompt file has no emoji, arrows or symbol characters
- Brand and headline read correctly letter by letter, with key text and faces inside a 6% margin on every side
- Every source name, logo, slogan, person, interface and watermark is replaced by the user's own identity or a new one, and the layout is new for this subject
- Exactly one paid request per approved generation
What you need before you start
- You bring
- Brand or project name, spelled as it should appear on the card · One short headline · What the page, product or audience is · Optional: brand colors, a logo the user may use, a mood, owned reference images, a platform safe zone
- Dependencies
- Node.js 20+ to run the included FAL script · Your own FAL API key, set as FAL_KEY; FAL charges for each image
- Outside costs
- FAL_KEY
- Last tested
- v0.1.0 — Claude Code / Codex / Cursor
User reviews
Loading reviews…
FAQ
My brand color isn't blue. Does the style still work?
Yes. Send your main color with the name and headline (and a logo, if you may use one). The Skill keeps the style's color relationships, not its exact values, and redraws the lines around you; no outside mark is copied.
Can it work from an image I already have?
If the image is yours or you may use it, yes. Hand it over with --reference and GPT Image 2 runs in edit mode instead. Only mood, color and layout are taken from the picture; nothing in it is reproduced on the card.
Is there a cost beyond the Skill?
Yes: FAL bills every image to your own account. The helper never retries by itself, and running the same command again only picks up the saved request, so checking on a slow job never pays for a second one.








