AI design bad.

Most AI-generated interfaces look the same. Same layout. Same spacing. Same nav bar. Same everything. You've seen it a thousand times and you'll see it a thousand more, because every AI is trained to make "good enough" design, not good design.

Good design exists. Good design looks a particular way. It has rules, whether or not anyone wrote them down.

Horus looks at good design. Horus finds the rules. Horus gives those rules to your AI.

Claude. Lovable. v0. Whatever you're using.

You ask your AI to design a page. Your AI calls Horus. Horus tells your AI: design it the particular way good design is.

That's it. That's the whole idea.

npm: horus-codec

The problem

AI design tools are trained on the average of everything, so they produce the average of everything. Generic spacing. Generic hierarchy. Generic nav. It's not wrong, exactly. It's just nothing.

Real design isn't random. A clean interface follows patterns: how elements are grouped, how strict the spacing is, what shape things take, when the rules bend and when they don't. Those patterns are learnable. Most AI tools just never learned them.

What Horus does

1. Show Horus real design. Feed it reference images of an interface style you actually like.
2. Horus extracts the rules. Not just colors. Structure. Hierarchy. Spacing. Grouping. The stuff that actually makes a design feel intentional instead of generic.
3. Horus writes a ruleset. A reusable definition of that visual language, saved under an ID you choose.
4. Your AI asks Horus for a spec. Give it a brief and a ruleset ID. Horus returns a structured UI spec built from those rules.
5. Your AI builds the interface. Same brief, same AI, different result, because now it's not guessing.

Brief → LLM → Horus → UI Spec → LLM → Interface

It's not guessing, it's grounded