OpenAI launches Canvas, a multimodal model for responsive UI generation

AI · 5 min read

Canvas combines text, rough sketch input, and existing screenshots to produce adaptive, code-first layouts. The model outputs both high-fidelity visual mockups and structured layout metadata such as grids, constraints, and CSS-like tokens that can be exported to popular design tools.

OpenAI highlighted on-device runtime options for basic inference and a cloud service for heavier synthesis, with a focus on privacy for enterprise customers. Early tests by design teams showed Canvas reduced initial layout iteration time by up to 60 percent when used to generate baseline pages and design variants.

OpenAI is also releasing a plugin SDK so Canvas can be integrated into tools like Figma, Framer, and IDEs. The company emphasized model alignment work around accessibility and responsive behavior to avoid producing inaccessible patterns by default.