Canva has introduced its first proprietary AI image generator, branded as Canva Craft, positioning the tool against offerings from OpenAI, Microsoft and Google. According to the company's Thursday announcement, the model employs a two-stage architecture that first outlines a design before producing the final visual output. Canva says this approach delivers higher precision, consistency and user control over generated content.
Craft can render images at resolutions up to 2,048 pixels on the longer side and supports more than thirty distinct aspect ratios. The service is marketed for creating social media graphics, marketing assets, realistic scenes, digital artwork, and environmental visuals. Canva highlights that the two-stage pipeline improves generation of complex compositions that require legible typography and exact placement of multiple objects within a single frame.
In April, Canva added prompt-based editing tools that let users describe desired changes to a design, enabling AI-driven modifications through natural language. The platform also introduced layer support for AI-generated images, allowing selective adjustments to portions of a picture while preserving the rest. These features lay the groundwork for Craft’s more sophisticated planning capabilities, extending the creative workflow beyond simple image synthesis.
Canva reports that Craft was trained on millions of human-crafted templates drawn from its extensive content library, which includes social media posts, presentations, infographics and menus. By leveraging this curated dataset, the model learns typical layout structures, color schemes and typographic conventions used across a wide range of commercial design assets. The company claims this exposure helps the AI produce outputs that align closely with professional design standards.
On the crowdsourced Design Arena leaderboard, which benchmarks AI design tools, Canva Craft currently holds the fourth position. It trails behind OpenAI’s GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT-Image-2 models. The ranking reflects Craft’s competitive performance while also indicating room for improvement relative to the leading offerings from OpenAI in terms of image quality and versatility.
Canva emphasizes that Craft incorporates robust moderation, privacy safeguards and reporting mechanisms to address potential misuse. The company also plans to expose the model to external AI agents via the open Model Context Protocol, and it has announced upcoming ‘Craft-fast’ and ‘Craft-ultrafast’ variants designed for quicker inference. The service is live worldwide and can be accessed through the Dream Lab option in Canva’s left-hand navigation menu.