OpenLayer's Aurora diffusion v4 targets UI asset generation with native component metadata

AI · 6 min read

Aurora v4 outputs PNG or SVG files with an accompanying JSON sidecar that maps shapes to component names, tokens, and constraint metadata. This metadata-aware output allows designers to import generated assets into tools like Figma or Sketch and retain editable layers and styles.

The model was trained on paired design files and bitmap renderings to learn correspondences between visual forms and component semantics. OpenLayer provides an API and a plugin that automatically converts sidecars into native layer trees or suggestions for mapping to a project's design tokens.

OpenLayer also introduced style-control prompts and a strict-mode that enforces token usage during generation, aiming to keep generated assets production-ready. The company sees Aurora v4 as part of a broader shift toward generation outputs that are directly composable, not just inspirational.