Canva is making its first image model available to users worldwide and developers, with a planning stage that organizes a composition before rendering. The company said users worldwide can access Craft in Dream Lab, and developers can use the public API.
Craft divides image creation into planning and rendering. Canva says the planning stage sets visual relationships within a composition and determines text placement before the system generates the image. Outputs can reach 2K resolution and come in more than 30 aspect ratios, the company said.
Canva is positioning Craft for graphic work that combines objects and lettering. The model is designed to produce legible text and accurately positioned objects. Canva says its architecture allows more targeted changes to individual elements while keeping the rest of a composition intact; more advanced editing features are planned.
Training drew on millions of human-created templates in Canva’s design library, including posters, social media posts, presentations, infographics, menus and typography-heavy layouts. The examples came from designed materials rather than photographs. Images made with Craft can also be converted into editable elements through Magic Layers, Canva’s image-to-design tool.
Canva linked the project to the computing demands of operating AI tools at its scale, citing the need for efficiency and faster inference. It said model distillation and a custom inference engine support development of two faster versions intended to reduce generation costs. Those variants are still in development.
Canva also says its Canva MCP Server is coming soon to connect Craft with AI assistants and workflows. The company has not announced release dates for the server or faster versions.
