OpenAI launches Canvas: a multimodal model tuned for designers

AI · 5 min read

OpenAI’s Canvas is positioned as a bridge between generative AI and practical design workflows, combining image, vector, and layout reasoning in a single model. The company says Canvas was trained on paired UI screenshots, design tokens, and code snippets so it can propose accessible color palettes, layout hierarchies, and export-ready SVGs.

In demos, designers could sketch wireframes and ask Canvas to generate responsive variants, convert raster assets to vector, or produce component code for React and SwiftUI. A key selling point is layout awareness: Canvas attempts to preserve responsive constraints and accessibility properties when transforming or generating assets.

OpenAI will distribute Canvas via an API and a Figma plugin in beta next month, with on-device, privacy-first runtimes planned later in the year. Pricing and enterprise licensing were not finalized, but OpenAI emphasized partnerships with design platforms and tool vendors to reduce workflow friction.