Canva has launched its proprietary image generation model, Canva Craft, designed to compete with solutions from OpenAI, Google, and Microsoft. The model utilizes a two-stage architecture that plans design layouts before generating final images, which improves the accuracy of text placement and object positioning.
The new tool supports image generation at up to 2K resolution and offers over 30 different aspect ratios. It is capable of creating various types of content, including marketing materials, social media posts, digital art, and realistic environmental scenes.
The model was trained on millions of human-made templates from the Canva library, such as menus, infographics, and presentations. Currently, Canva Craft holds the fourth position on the Design Arena leaderboard, trailing behind three models from OpenAI.
Canva has integrated safety features into the model, including input and output moderation, reporting tools, and privacy controls. The platform intends to support AI agents via the open Model Context Protocol and plans to release faster versions, specifically Craft-fast and Craft-ultrafast.
The tool is now available to all global users. Access is provided through the Dream Lab option located within the three-dot menu on the Canva interface.
