ZendaSketch: new vector-aware image model built for UI prototyping

AI · 5 min read

ZendaSketch differs from many image models by natively emitting structured vector output rather than raster PNGs, enabling designers to drop results directly into vector editors and component systems. The company trained the model on paired raster/vector datasets and added a proprietary 'component recognition' head to map artwork to library symbols.

In tests, ZendaSketch reduced manual cleanup time for rough wireframes by over 60 percent, according to Zenda's benchmarks, largely because the model preserves component anchors, spacing constraints, and exportable CSS variables. It also supports a 'snap-to-library' mode that replaces rendered UI elements with exact matches from a connected design system.

Zenda is releasing SDKs for Figma, Sketch, and a VS Code plugin for developers to call ZendaSketch during storybook generation. The immediate integration focus aims to shorten the loop between ideation, prototyping, and developer handoff.