The new feature and how it works
Canva has unveiled Canva Craft, its first image-generation model, designed to compete with those from OpenAI, Microsoft, and Google. Its distinctive element is the two-phase architecture: the system develops a plan for the graphic design before producing the image. According to Thursday's announcement as reported by The Verge, this approach ensures greater precision, consistency, and control.
The model generates images at up to 2K resolution and in more than 30 formats. In addition to social media posts and marketing content, it is designed for realistic scenes, digital art, and environmental images. Upfront planning is especially useful in complex cases, where legible text and objects placed in precise positions are needed.
Training and positioning
Canva says it trained Craft on millions of templates created by people, drawn from the platform's library: social posts, presentations, infographics, and menus. The model currently ranks fourth in the Design Arena leaderboard, the crowdsourced benchmark for image-generation models. Ahead of it are three versions of OpenAI's GPT-Image: GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT-Image-2.
The launch is part of a broader strategy. In April, Canva had introduced prompt-based editing features that allow users to create or modify a design by describing it, and had added layers to AI-generated images, so that one part could be edited without affecting the rest. Craft therefore represents a further step: not only modifying, but generating from scratch with a claimed control over structure.
Availability, safety, and open questions
Canva states that Craft includes robust moderation of inputs and outputs, privacy and security controls, and reporting tools. The company plans to allow AI agents to connect to the model via the Model Context Protocol, the open standard for integration, and to launch more efficient versions called Craft-fast and Craft-ultrafast in the future.
The model is available immediately to all users worldwide, accessible from the three options in Canva's left-hand menu and then from Dream Lab. Some aspects remain to be clarified: the material does not indicate the commercial terms of access, the technical details of the training, or how copyright issues relating to the templates used are handled. The fourth-place ranking in Design Arena is a company statement based on a crowdsourced benchmark, not an independent measurement. Real-world performance on complex text and precise placements will need to be verified through use.
