# Picbreeder Design

A design-breeding run can end with a family, not a champion.

Picbreeder made this easy to see. It gave people populations of generated images, let them select the ones that appealed, and used those selections as parents for the next generation. Published images could become branches that other people continued. The user did not need a metric for beauty. A click carried enough taste to guide evolution.

Interface design can use the same shape. Generate a population. Let taste choose the child with the living effect. Preserve that effect and reopen the rest of the design space. If the pull was a depth illusion in blurred text, the next children should test whether the depth came from blur, motion, layout, timing, typography, or the relation between a control and the atmosphere around it.

Early in the run, taste chooses a direction. One child has the field. Another has the motion. Another has the quiet center. The feedback says: keep this gene and mutate everything around it.

Later, the signal changes. The children stop looking like separate worlds. They begin to look like nearby points in one space. The question shifts from *what is this?* to *how much motion, density, contrast, latency, and semantic weight can this carry?*

That is dimensionality reduction by taste. The evaluator never wrote an objective function. She only kept saying where the life was. The population absorbed those choices and shed whole categories of variation. What remains is a smaller space the team can build inside.

This is the moment a normal review process often makes its worst move. It asks for the winner.

A single winning mockup is too lossy after convergence. It hides the axes that still matter. The family preserves them. One child shows the quiet end of the motion range. One shows the dense end. One shows how far color can move before the center loses stability. One shows the maximum atmosphere the page can carry before the call to action dissolves. Together they form a spec with the knobs still attached.

This matters most before metrics exist. A landing page, an onboarding flow, or a first-run interaction has to earn its first click before analytics can say anything trustworthy. At that stage, taste is the instrument that arrives on time.

A Picbreeder design run should stop when another generation would mostly polish. The children share a grammar. The remaining differences can be named as implementation parameters. The reviewer no longer says, "that one." She says, "this is the template."

That sentence is the stop signal. The output is the smallest design space that still contains the living thing.

**Source:** Jimmy Secretan et al., [*Picbreeder: Evolving Pictures Collaboratively Online*](https://campbellssite.com/papers/secretan_chi08.pdf) (CHI 2008); [Picbreeder](https://picbreeder.net/).
