Generative Imagery

Design visual rules that produce coherent image variations while keeping authorship, constraints, and review visible.

Generative Imagery

In brief

Generative imagery uses a defined system to make many related visuals. The designer shapes the rules, input, limits, and selection rather than drawing each final image separately.

A set of related sculptural variations made from shared shapes beside a selected output

Common tools

  • Adobe Illustrator
  • Figma
  • Browser tools

Skills in play

  • Concept development
  • Systems thinking
  • User testing

What it is

Generative imagery is made through rules that produce visual variation. A designer may define shapes, relationships, colours, data inputs, and limits, then select or refine the outputs. A repeating pattern is a simple rule-based system; a more complex system may generate a different composition for each person or event. Generative work is not automatically artificial intelligence. Many systems use explicit geometry or code without a trained image model.

Richard The and E Roon Kang’s MIT Media Lab identity is a useful example. Their visual system allowed individual marks to vary while preserving a shared construction. The value was not a pile of unrelated logos; it was a designed relationship between difference and recognition. The project also included interfaces for choosing marks and applications across media, showing that the generator was only one part of the identity.

Define the reason for variation

Ask what changes and why. Variation might express membership, location, time, data, or a controlled editorial choice. If nothing meaningful varies, a fixed illustration may be clearer and cheaper to maintain. Write down the invariant elements that make outputs recognisably related and the variables that may change. Set limits before production: allowed palettes, minimum contrast, density, scale, and forbidden combinations.

Choose trustworthy inputs

If the system uses data, identify its source, update frequency, gaps, and meaning. A striking image can imply that its shapes encode something precise even when the input is arbitrary. Explain the relationship honestly. For example, if a colour is randomly assigned, do not describe it as a portrait of the user’s personality. Keep a way to reproduce an output when it needs correction or approval.

For systems using a trained image model, additionally review source permissions, output rights, model behaviour, and whether generated details could be mistaken for factual evidence. These questions belong in the brief, not as an afterthought once a visual has been selected.

Design the system and its boundaries

Begin with a small set of simple rules and generate many examples. Look for collisions, empty compositions, distracting clusters, and outputs that lose the core idea. Test extreme input values, long names, small display sizes, and monochrome reproduction. A system that yields one beautiful sample but many weak ones needs better constraints.

Create a selection rule. Will every output be used, will an editor approve them, or will users choose from a bounded set? Document the decision because it determines workload and responsibility. Visual identity systems need practical application rules; generative variation does not remove the need for art direction.

Keep the output accessible and usable

Test contrast and visual hierarchy in real contexts. A dense composition behind a headline may make the text hard to read. A rapidly changing visual may distract or create motion barriers. Provide a fixed fallback and descriptive alternative text when the generated image carries meaning. If the output is decorative, avoid presenting its internal parameters as essential information.

Prototype in context

Place a range of outputs into cards, headers, motion frames, print applications, and social crops. Check whether they still feel like a family and whether a viewer can identify the important information. Ask another designer to produce a new output from the documented system. If they cannot, the rules are living only in the original designer’s head.

Keep a record of version, input, selected output, and any manual edits. The ability to explain and recreate an image is part of the craft, particularly when many people will publish from the same system.

Practical checks

  • What exactly varies, and what meaning does that variation carry?
  • Do the rules prevent weak or unusable outputs at the extremes?
  • Can an output be reproduced, reviewed, and corrected?
  • Does the visual remain recognisable and readable in real applications?

Common mistake

Confusing abundance with a system. Thousands of outputs are useful only when their differences serve an idea and the weakest examples still work.

From our kitchen Take a look at graphic recipes from our chefs πŸ₯‘

Sister pages

Explore Illustration and Image-making

Let’s Dish It Out

Send us your brief, your wildest idea, or just a hello. We’ll season it with curiosity and serve back something fresh, cooked with care.

Talk to the Chef πŸ₯‘