Generative identity for SaaS and AI brands

Most brand identities are a set of files: a logo, a palette, a few layouts. A generative identity is a set of instructions. You define the parts and the rules, and code produces the visuals, so the brand can respond to what is happening in the product instead of sitting still on top of it. This guide covers what that means, when it fits a SaaS or AI brand, when it doesn’t, and how to keep it from turning into noise.

What makes an identity generative?

A fixed set of inputs (shapes, grids, type, color) and rules for combining them, run by code. The output varies, the system doesn’t.

In practice a generative identity has four parts:

• 01Inputs: the raw material, like a grid, a set of shapes, a typeface and a palette.
• 02Rules: how the inputs combine, what is allowed to vary and by how much.
• 03A driver: what makes the output change, such as data, time, a user action or a seed.
• 04Code that runs it, in the product, on the site or in a tool that exports assets.

A logo in five colors is a set of variations, not a generative identity. The test is whether someone can get a new, correct output without a designer drawing it, because the system knows what correct means.

The output varies. The system doesn’t.

When does it fit a SaaS or AI brand?

• 01The product generates, transforms or visualises data. A brand that changes with input says something true about a product that does the same.
• 02You need lots of on-brand assets: social, docs, product empty states. A system that generates them keeps up with the volume without a designer in the loop for each one.
• 03You want to avoid the sameness of AI branding. A lot of AI products share the same look, and a visual language built on your own rules is harder to confuse with the next one.

When doesn’t it fit?

When the team can’t maintain code-based assets, or when a simple, fixed mark would do the job.

A generative system is software, and software needs an owner. If nobody on the team will keep it running, it will quietly be replaced by screenshots of old outputs. It also comes second, not first. Most SaaS and AI products need the basics before anything moves: type, color, spacing, logo rules and voice, applied everywhere. We cover those in Systems outlive logos. A generative layer works best on top of that, not instead of it.

How does it work across product, website and social?

The same rules run in different places, at different intensities:

• 01Product: empty states, loading moments and workspace covers generated from the user’s own data or IDs, so each account looks like itself and still like the brand.
• 02Website: a live hero or background that reacts to the pointer or to real product data, with a static version for slow connections and reduced motion.
• 03Social and docs: templates that produce a new graphic from a seed or a title, so every post and page preview is on-brand without being drawn by hand.

The product usually wants the quietest version. People are there to work, so the generative layer should support the task, not compete with it.

How do you keep it consistent?

Write the rules down, limit the parameters, and ship a static fallback for small sizes and favicons. Use the same tokens in the product.

• 01Write the rules in plain language next to the code, with examples of good and bad outputs.
• 02Limit the parameters. Set ranges for every value that can change, and keep the ranges narrow.
• 03Keep some things fixed. The wordmark, the typeface and the core palette usually shouldn’t move.
• 04Ship static fallbacks for favicons, app icons, small sizes, print and anyone who prefers reduced motion.
• 05Use the same design tokens in the generator, the product and the site, so color and type never drift between them.
• 06Set a performance budget. A brand animation that slows the page down costs more than it adds.

Examples from our Playground

Our Playground is where we test techniques in code before using them in real work. Three of the experiments show the kind of building blocks a generative identity is made from. They are studies, not finished brands.

ASCII flow: a rotating solid drawn live from a luminance ramp. A texture system that could become a brand pattern.

Particle trail: a cursor-driven trail of particles that drift and fade. A motion principle for interaction moments.

Grid warp: a dot grid that bends around the pointer. A structural motif built from a grid.

How do you start?

• 01Get the fixed basics right first: type, color, spacing, logo rules and voice.
• 02Pick one generative element, tied to something the product actually does.
• 03Prototype it in code early, because the rules only become clear once they run.
• 04Decide the ranges and the fixed parts, then build the static fallbacks.
• 05Document it in the guidelines and in a plain-text file your team and its AI tools can read.

Brand identity with us starts at US$6k for the basic package; a generative layer is a bigger scope and costs more. What a brand identity actually costs explains what the basic package covers, and the services page has the rest.

If you want a brand that moves with your product, send your brief.

• Identity
• Systems
• Generative