Canva Launches Its First Image-Generation Model for Design Work

Canva launched Canva Craft, its first image-generation model, for all platform users, introducing a two-step process that maps a composition before rendering the final image.

Users can access Craft through the Dream Lab tab. Canva plans to make the model available through its API and add image generation to third-party chatbots through its Model Context Protocol (MCP), including ChatGPT and Gemini.

Craft first turns a prompt into a plan for the image. That plan covers where elements go, their relative scale, their visual order and information for generating text. The model then renders an image based on the plan.

Craft can produce images at resolutions up to 2K and in as many as 30 aspect ratios for different platforms. An aspect ratio determines an image’s shape, giving users options for different formats.

The model is designed to create detailed objects and readable text. Targeted editing is planned to let users change one part of an image while keeping the other areas intact.

Rather than photographs, Craft’s training material came from designs in Canva’s library. The examples include posters, social media posts, presentations, infographics, menus and layouts with substantial typography. Canva’s focus on designed compositions is reflected in the types of material used to train the model.

Images made with Craft can be converted into layered designs using Canva Magic, the company’s image-to-design tool. The conversion gives users a layered design based on the generated image.

Canva is developing two lower-cost versions, Craftfast and Craft-ultrafast, which it plans to release soon. Craft also gives Canva more control over quality, speed and costs.

Researchers from Leonardo.ai contributed to Craft. Canva acquired the media-generation platform in 2024.