Google DeepMind Launches Layout-GEN: A Benchmark-Tuned Model for Responsive UIs

AI · 5 min read

Google DeepMind has released Layout-GEN, a model trained on layout-annotated datasets and a range of style constraints to produce responsive UI templates. The model outputs structured layout trees, CSS grid configurations, and accessibility overlays that highlight contrast and semantic roles.

Accompanying Layout-GEN is the LayoutBench benchmark, which evaluates generated interfaces for responsiveness, content preservation, and accessibility adherence across desktop, tablet, and mobile viewports. DeepMind said LayoutBench will be open-source to encourage reproducible comparisons and safer generative UI research.

DeepMind also published a research paper detailing a compositional training approach that separately models content placement and style application, which improves transfer to unseen component libraries. The model API is initially available to select partners before broader availability later this year.