Canva Research, the research division of Canva, has developed and trained Craft, the company's first image model, entirely in-house from scratch. Craft is available to all Canva users in Dream Lab, the image generator in the left sidebar of the editor, including those with a free account. Developers can also access the model via the Canva interface, as Canva announced in the developer forum.
Until now, Dream Lab ran on Phoenix, an image model from Leonardo.ai, the subsidiary acquired in 2024. According to TechCrunch, Canva has also used several models from other providers and is likely to save costs with its own model. With it, Canva wants to better control quality and speed.
What does Craft do differently?
Craft works in two stages. In the first step, a blueprint is created: where objects are placed, how large they are in relation to each other, what takes priority in the image, and what text goes into it. Only then does Craft render the image, in up to 2K, meaning around 2,000 pixels on the long side, and in up to 30 aspect ratios. Canva expects this process to deliver legible text and precise details. Individual elements should also be changeable without the rest of the image shifting.
Canva Research trained the model mainly on templates from its own library rather than on photos, including posters, social media posts, presentations, infographics, menus, and text-heavy layouts. Craft accepts prompts in any language, including German, and the model renders text in the image in 55 languages.
What does a Craft image cost?
Canva does not state a separate price for Craft. Via the interface, each Craft image costs AI credits from the user account, just like an image from the other models. According to TechTimes, the company has not yet specified whether Canva counts Craft as standard or premium usage within your plan's allowance.
Craft is not fast. For an image via the interface, Canva estimates 50 to 60 seconds, because the planning step runs before rendering. According to TechCrunch, two cheaper and faster variants are in the works. Craft is also expected to be usable soon via Canva's MCP server, the connection for AI assistants such as ChatGPT and Gemini.
What this means for your graphics
Craft becomes especially interesting for posters, invitations, and social media graphics with a headline. It's best to try the model on a real draft with a German headline and check umlauts and spelling in the image before the graphic goes to clients.
For photorealistic product images, on the other hand, it's worth comparing it with your previous results, since photos were not the focus of the training.
If only one detail doesn't fit, you convert the image with Magic Layers into a design with individual layers and change the text or color there. That way, you don't have to generate a new image for every correction. How to design and edit graphics in Canva is shown in the Canva course for graphics, images, and videos.
If you use the Canva interface for your own workflows, it's best to plan fixed production processes with Craft only later. Canva has so far provided the interface only as a preview version and points out that changes can render existing integrations unusable.
Canva introduced Craft on October 22, 2026.
Sources
