OpenWeave 2.0: an open-source multimodal model for UI extraction and prototyping

AI · 5 min read

OpenWeave 2.0 improves UI understanding by combining vision transformers with a structural decoder that outputs labeled component trees and CSS-like layout hints. The project aims to create a shared format for converting screenshots into editable prototypes across tooling ecosystems.

The model supports a 'confidence map' for each detected component, enabling designers to quickly review low-confidence areas. OpenWeave also ships with conversion scripts for Figma, Sketch, and HTML/CSS to speed prototyping workflows. Contributors highlight the value of an open model for research and tooling interoperability.

Because it's open-source, companies can adapt OpenWeave for custom component libraries or accessibility audits. The community around the project is already producing training datasets for internationalized UIs and right-to-left layouts, addressing gaps seen in earlier closed models.