Design-First AI Architecture
Unlike standard image generators that prioritize photographic scenes, Craft was trained on millions of human-designed templates, including social graphics, presentations, and menus. The model utilizes a two-stage process: first, it interprets a prompt to establish a plan for visual hierarchy, object placement, and typography; second, it renders the final image based on that structural blueprint. This approach is intended to minimize the common AI issues of broken text and awkward layouts.
A key feature for marketers is the model's integration with Canva’s "Magic Layers," which allows users to convert generated images into editable design elements rather than flattened files. This functionality is intended to streamline the revision process, enabling teams to resize assets, swap offers, or translate copy without needing to regenerate the entire image.
Availability and Future Roadmap
Craft is currently available to global users via the Dream Lab interface and is accessible to developers through Canva’s public API. While the company has teased future "Craft-fast" and "Craft-ultrafast" variants, these are not yet available. Additionally, while the model is positioned to support third-party AI assistants, full integration is still in development.
As Canva moves to own more of its AI infrastructure, the company aims to reduce latency and improve control over design workflows. However, users are encouraged to test the model against specific brand requirements—such as multi-format campaigns and regional variations—to determine how effectively it handles complex, iterative design tasks.
