From prompt to production

Getting started

Open any layer, click Copy prompt and paste it into the tool that writes your code. Nothing needs installing for the basic flow.

The prompts are written in Spanish, and the English version adds a line asking your agent to build all visible copy in English. The default output is a self-contained HTML file: open it in the browser and it's already running, motion included.

Which model to use

Every layer is written and checked with Claude Opus 5, which holds design and motion language best across a long prompt. Several pieces have also been run with OpenAI's Codex and with a local model, and are judged with the same checks.

Any agent that accepts a prompt of several thousand tokens can handle them. If the result falls short, the model has almost always truncated its answer: ask it to continue.

Bring it into your stack

The prompt doesn't mention any framework unless you ask it to. Add one line at the end and your agent will build straight into yours.

Build it as a Next.js 16 component with the App Router and Tailwind v4.
Use my design tokens from src/app/globals.css instead of literal colours.
The component goes in src/components/sections/ and must not use 'use client'
unless it needs state or effects.

Fine-tuning the result

Change the content and colours first; the motion gets tuned last. Polishing the animation before you have the real copy is wasted time.

If something feels off, describe it as a feeling, not as code: “the entrance is too abrupt, let it breathe” works better than asking for a specific easing value.

License

Any paid plan gives you a commercial license: no attribution, no limit on sites, and client work included.

What you can't do is resell the prompts as they are, redistribute the library, or automate bulk downloads of it.