Figma explains how its AI tool copied Apple’s design

To give the model enough freedom to compose designs from a wide range of domains, we commissioned two extensive design systems (one for mobile and one for desktop) consisting of hundreds of components, as well as examples of different ways to combine those components that were used to guide the results.

We feed the metadata from these handcrafted components and examples into the model’s context window, along with a user-input prompt describing the design goals. The model then efficiently assembles a subset of these components, inspired by the examples, into fully parameterized designs. Amazon Titan, the diffusion model, then creates the images needed for the design. This is about as elementary as helping the AI ​​identify, arrange, populate, and theme diminutive, collapsible templates from the design system to provide a starting point.