Canva has released Canva Craft, its first internally developed image-generation model. The system is available to all users through Dream Lab and is planned for the Canva API and third-party assistants through the company’s Model Context Protocol integration.
Craft separates generation into two stages. It first plans spatial relationships, relative size, visual hierarchy and text, then renders an image at up to 2K resolution in one of 30 aspect ratios. Canva says the structure improves legible text and allows a user to edit one area without unintentionally changing the rest.
Training emphasized human-designed templates from Canva’s library, including posters, presentations, menus, social posts and infographics, rather than relying only on photographs. Output can be passed through Canva Magic, the company’s image-to-design model, to create a layered design whose elements remain editable.
Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to the model. Canva says owning the generation stack gives it more control over quality, speed and API costs. Two faster versions, Craftfast and Craft-ultrafast, are still under development, and the company has not provided dates or independent quality comparisons for them.