Canva launches Craft, its first in-house image model, which plans a design's layout and text before it draws.

On Thursday, Oct 22, Canva released Canva Craft, an image generation model built and trained from scratch by its own research lab (Canva). Most image models turn a prompt straight into pixels. Craft works in two steps: first a planning model writes out the layout, meaning which elements go where, how big they are and what text should appear, and then a second model draws the image from that plan. Canva says this helps with readable text and precise placement, and lets you change one element without redoing the rest. It makes images up to 2K resolution in more than 30 aspect ratios, takes prompts in any language and can write text in 55 languages.

Canva trained it with an emphasis on design, using millions of human-made templates from its library, such as posters, social posts, menus and infographics (TechCrunch).

Craft is available now to all Canva users through Dream Lab, and to developers through Canva's API, where Canva's developer notes say a job takes about 50 to 60 seconds and uses your AI credits (Canva Developers).

Support through Canva's MCP server, so AI assistants can use it, is coming soon. The Verge notes that Craft is fourth on the Design Arena leaderboard, a crowd-voted ranking of AI design images, behind three OpenAI image models (The Verge).

Canva says faster, cheaper versions called Craft-fast and Craft-ultrafast are on the way.

Try it: open Dream Lab in Canva and ask for a poster or social post with a headline. Check whether the text comes out spelled right, then try Magic Layers to see how easily the result splits into parts you can edit.