Google Vertex AI introduces Figma-to-Model pipeline for designers

Tech · 6 min read

The pipeline extracts structured tokens, layout hierarchies, and annotated components from Figma files, automatically generating labeled pairs for tasks such as copy replacement, accessibility fixes, and adaptive layouts. Vertex AI then offers automated training recipes tuned for these design tasks.

Google emphasized privacy and governance: extracted datasets can remain in a team's private VPC, and Vertex AI includes dataset diffs and annotation provenance so designers can audit training inputs. The product also supports differential privacy options during training.

Designers can iterate quickly with a 'train-and-preview' loop in Vertex where model outputs are rendered back into Figma frames for evaluation. Early partners reported faster automation of repetitive tasks and improved handoff between design and engineering.