Canva has officially launched Canva Craft, its first homegrown image generation model developed and trained by Canva Research. The model enters the market with strong independent results, debuting as the fourth most advanced image generation system on DesignArena's leaderboard and making Canva Research the world's second most advanced lab for image generation behind OpenAI. Available now inside Canva and through a public API, Craft is positioned as a design-first foundation model for creators, developers, and product teams.
A Two-Stage Architecture Built for Design
Craft uses a two-stage architecture that separates compositional planning from image rendering, unlike conventional models that generate pixels directly from a prompt. In the first stage, an image planning model interprets the prompt and writes a detailed plan of elements, spatial relationships, relative scale, visual hierarchy, and any required text. A dedicated rendering model then translates this plan into a high-fidelity image at up to 2K resolution across more than 30 aspect ratios.
Benchmark Results and Competitive Ranking
On DesignArena's independent leaderboard, Craft ranked fourth overall and achieved first place globally for People & Portraits and Landscapes & Nature. It also secured second place for Abstract & Patterns and fourth for Products, placing ahead of models developed by Google, Microsoft, Meta, xAI, and ByteDance. Canva said the preliminary results prompted the company to bring forward its announcement, underscoring that a relatively small lab can compete with the world's best-funded AI developers.
Efficiency and Future Model Variants
Canva said delivering frontier-level performance at scale required breakthroughs in compute efficiency and inference speed. Through advanced model distillation, the company reduced the number of inference steps required by Craft to a fraction of the original model, while a custom inference engine optimized the planning stage. These advances underpin Craft-fast and Craft-ultrafast, two upcoming variants designed to deliver high-quality image generation at a fraction of the cost of leading frontier models.
Training on Design and Designer Feedback
Craft's training data emphasizes design far more than typical image models, drawing on millions of human-made templates from Canva's content library across posters, social posts, presentations, infographics, menus, and typography-rich layouts. Following pre-training, Canva optimized the model through supervised fine-tuning and reinforcement learning using feedback from professional designers. For internal evaluations, designers and researchers assessed more than 25,000 pairs of images, and when automated judgments differed from human judgments, Canva sided with the designer.
Availability, Developer Access, and Safety
All Canva users globally can access Craft starting today through Dream Lab in the left panel of the editor. Developers and product teams can also build Craft into their own tools through a public API, while Canva MCP connectivity for AI assistants and automated workflows is coming soon. Craft is backed by Canva Shield, the company's comprehensive AI safety framework that includes input and output moderation, privacy and safety controls, and reporting tools for potentially harmful or misleading content.
The launch of Canva Craft marks a material step in Canva's strategy to develop its own foundational AI technology, moving beyond reliance on third-party image models. With strong benchmark performance, design-specific training, and a focus on compute efficiency, Canva is building a credible in-house platform that can serve both everyday creators and enterprise developers. As faster Craft variants and expanded editing capabilities arrive, the company is signaling that world-class AI research can emerge from outside Silicon Valley.