Canva announced on Thursday the launch of its image generation model Canva Craft, officially joining the AI image generation competition with OpenAI, Microsoft, and Google. The model adopts a two-stage architecture that can plan the design before generating an image, improving the precision, consistency, and controllability of image generation.
Canva Craft supports image generation at up to 2K resolution and aspect ratios exceeding 30, with use cases spanning social media content, marketing materials, digital art, and environmental imagery. Its two-stage generation mechanism is optimized for complex designs containing clear text and precise object layouts, and can handle element positioning and composition more accurately.
The model was trained on millions of human-designed templates from Canva's content library, covering social media posts, presentations, infographics, menus, and more. In April this year, Canva already launched prompt-based design editing, supporting the creation or modification of designs through natural language, and introduced layer editing for AI-generated images, allowing users to make local adjustments without affecting other areas.
In the Design Arena leaderboard of AI image generation models, Canva Craft currently ranks fourth, behind OpenAI's GPT-Image-2.5Sunburst, GPT-Image-2.5Flare, and GPT-Image-2. Canva said the model comes with input and output moderation, privacy and security controls, and content reporting tools, and plans to support AI agents accessing it through the Model Context Protocol (MCP), with faster Craft Fast and Craft Ultra Fast versions to be launched in the future. Currently, Canva Craft is available to users worldwide through Canva's Dream Lab