Canva Craft plans image layouts before drawing them
Canva has released Craft, its first internally developed image-generation model. The system plans the arrangement of text and objects before rendering a picture, drawing on design templates in the company’s library. It connects image generation with targeted changes and layered editing, while faster variants and additional connections remain on the development roadmap.
Artificial Intelligence··Morning
Craft plans the page before rendering
Canva has released Craft, the first image-generation model developed by its own research team. The design platform introduced it on October 22, adding an internally built model to the external image models it has already used. Craft is available inside Dream Lab. Its distinguishing mechanism separates planning a composition from rendering the finished image: the model first interprets the instruction and establishes where the parts should go.[1], [2]
The plan includes the spatial relationship between objects, their relative sizes, the visual hierarchy and the text that should appear. Rendering then turns that arrangement into an image. Canva describes this architecture as supporting readable lettering and more precise placement. Those descriptions come from the company; the release does not supply an independent measurement establishing the same result for every design task.[1]
Training draws on designed layouts
Craft’s training material includes human-designed templates from Canva’s library. Posters, social posts, presentations, infographics, menus and layouts with substantial typography give that material a design focus beyond photographs. The release connects the model to work already done inside the platform: arranging text and pictures within a finished composition. Researchers from Leonardo.ai, a company Canva previously acquired, contributed to development alongside Canva’s own research effort.[1]
Editing and faster variants remain separate tracks
The company emphasizes changing a selected part of an image without replacing the entire composition. Converting generated pictures into layered designs connects that output to Canva’s editing tools, where individual elements can be refined. The approach puts generation and subsequent adjustment within the same design workflow. It also distinguishes the model’s creation stage from the later work of turning a picture into an editable design.[1]
Further distribution through developer interfaces and connections to third-party chat applications is on Canva’s roadmap. The company is also developing Craft-fast and Craft-ultrafast as lower-cost variants for later release. Their names appear as planned products, rather than evidence that both have already shipped. Craft’s launch and the timetable for those variants are separate developments within the same model programme.[1]
