Generative branding: identities made by systems, not single images

Generative branding is an identity defined as a set of rules rather than a set of finished images. Designers decide the constraints, the colours, shapes, materials and limits, and a system, often a small piece of code, produces a new version of the logo or background each time one is needed. Every output is different, yet all of them are clearly the same brand, because they come from the same recipe.

How generative branding works

Every generative identity, from an art school's logo to a weather driven mark, has the same three parts. Once you can name them, you can design one.

The source matters more than it looks. A seed gives repeatable variety: the same number always produces the same result, so a design can be recreated exactly. Data ties the identity to something real, like weather or time. Format lets the system recompose for a story or a banner instead of cropping.

Examples of generative identities

• Visit Nordkyn, 2010: the tourism brand for a region in northern Norway uses a logo shaped by live wind direction and coloured by temperature, so it reflects the weather at that moment.
• MIT Media Lab, 2011: an algorithm produced a distinct logo for each member of the lab from a shared grid of overlapping shapes.
• OCAD University, 2011: a fixed black and white frame is filled with artwork by the university's own students, so the identity keeps renewing itself.
• Google Doodles: not code generated, but a long running system with one fixed rule, the name stays legible, and endless variation around it.

What they share is restraint. Each system varies one or two things, and each keeps something completely still: a frame, a grid, a name or a shape. The variety is what people notice; the constant is what they remember. These sit close to dynamic brand identity, which covers changing identities in general. Generative branding is the subset where the changes come from a system rather than a designer making each one by hand.

A generative background in a few lines of code

You can see the idea in a tiny example. The brand's constraints are a ground and a family of three violets. The variable is the position of the light, kept to the right of the frame so it never sits behind left aligned text. The source is a seed.

// A small seeded random generator: the same seed always gives the same numbers.
function mulberry32(seed) {
  return function () {
    seed = (seed + 0x6d2b79f5) | 0;
    let t = Math.imul(seed ^ (seed >>> 15), 1 | seed);
    t = (t + Math.imul(t ^ (t >>> 7), 61 | t)) ^ t;
    return ((t ^ (t >>> 14)) >>> 0) / 4294967296;
  };
}

// Constraints: these never change.
const brand = {
  ground: '#120d24',
  family: ['#4c1d95', '#7c3aed', '#c084fc'],
};

// Variables: light position and which tone leads, within limits.
function background(seed) {
  const rand = mulberry32(seed);
  const x = Math.round(60 + rand() * 30); // light stays on the right
  const y = Math.round(10 + rand() * 40);
  const lead = brand.family[1 + Math.floor(rand() * 2)];
  return `radial-gradient(circle at ${x}% ${y}%, ${lead} 0%, ${brand.family[0]} 40%, ${brand.ground} 80%)`;
}

document.body.style.background = background(2026);

How to write the rules for a generative brand

1. 1

   Start from a finished look

   Design one version you love by hand. The rules describe what makes it work, so you need a target first.

2. 2

   Separate fixed from free

   List what must never change. Usually the ground, the colour family, the material and the type.

3. 3

   Pick one or two variables

   Light position, leading tone or pattern density. Each extra variable multiplies the chance of an off brand result.

4. 4

   Protect the words

   Make legibility a rule: the busiest, brightest area must never land behind text. Text on a gradient explains why.

5. 5

   Review the extremes

   Generate many results and look at the strangest. If any feels wrong, narrow the range.

Generative branding is not AI image generation

The two get confused. An AI image model produces pictures from a prompt, and the same prompt can return very different results. A generative identity runs rules you wrote, so its range is known and a result can be reproduced from its seed. You can use AI to help design the rules or write the words, but the identity lives in the recipe. That also makes it easy to hand over: a new designer, an agency or an assistant can follow written rules far more reliably than they can imitate a mood board. AI design originality covers the wider question of where AI output comes from.

Marks are recipes

A Gradiently Mark works on the same principle. It is a recipe of colours, up to three materials, light, grain, motion and a type voice, rendered live rather than stored as a picture. It composes itself for every size you design in, and it reads where your words are so its calmest area sits behind them. Every Mark has a public code like GR·K7Q2·MX and exactly one owner at a time, so the recipe behind your look belongs to you. Mark materials explained shows what goes into one.