How it works
One definition of your brand’s taste, written once, obeyed everywhere.
Capture the judgement
We read what already exists — guidelines, tokens, shipped screens — and write it as a weighted brand definition: color discipline, type budgets, composition, voice, and everything your brand never does.
Translate it to any screen
The definition derives surfaces it has never seen: empty states, dense tables, error flows. Not a theme applied on top — rules composed from underneath.
Enforce it in every tool
Divona sits inside the AI tools your team builds with. Everything they generate flows through your definition before it ships, and drift gets caught when it happens.
The same prompt, twice.
Welcome back
Manage your account, view analytics, and explore new features from your dashboard.
Three drafts held overnight. Two are ready for your eye.
Every generation reviewed against your definition before you see it.
A definition, not a theme.
Five files your tools read directly. Every rule in them carries a weight, so a tool knows which ones it may bend under pressure and which ones end the argument.
What the brand is, and the anti-references it must never drift toward.
Colour, type, spacing, shape. The commodity layer — real values, measured rather than described.
How screens get built: loudness budgets, grids, density, the rhythm between blocks.
How it sounds in buttons, errors and empty states — and the words it refuses.
The negative space, carrying severity. The enforceable list of what the brand will not do.
The loudness budget
Each view gets exactly one loud object. Everything else must be quieter than it. If a view seems to need two loud things, the view is wrong — split it, or demote one.
Identity-critical. Never traded away. A tool that cannot satisfy a request without breaking one has to refuse the shortcut and solve it inside the system.
The default way this brand does things. Traded only when a core rule or a hard functional need forces it.
A starting preference. Adjusted freely when the surface genuinely demands something else.
Measured, not asserted.
Accessibility in a typical brand record is a sentence somebody typed, and nothing downstream can recompute it. Divona computes the ratio from the real values against the real surface — and fails a generation that breaks it.
computed from this page’s own tokens · wcag 2.2 · 4.5:1 normal text, 3:1 large and non-text
What a brand refuses is part of what it is.
Guidelines describe what to do. The drift comes from everything nobody wrote down. Divona writes the refusals as an enforceable list with severity, so the shortcut is closed before a tool reaches for it.
- coreNever a second hue. One accent, UI-wide — no gradients, no tinted panels, no supporting colour for variety.
- coreNever round a card, never square a button.
- coreNever two loud objects in one view.
- strongNever the slop kit: gradient text, glassmorphism, icon-topped feature-card grids, drop-shadow cards, purple palettes.
- coreNever hype copy. No exclamation marks, no supercharge, no unlock, no 10x.
the list above is divona’s own. this page obeys it.
Tokens are commodity. Judgement isn’t.
Every AI builder now accepts your colors and components. The output still looks generated, because a token file can’t say why your spacing feels calm, when your empty states stay quiet, or what your brand refuses to do under pressure.
Divona encodes that layer — weighted rules for composition, tone, and restraint — and makes it portable across every tool your team touches. Your judgement, shipping in places you’ve never looked.
Watch one judgement-level change reshape an entire product.
Start with your own brand — the setup takes the first two minutes.