Canva ships Craft, its first in-house image model

Canva, the Australian online design platform, released Canva Craft on October 22, 2026, the first image generation model that its in-house Canva Research lab built and trained from scratch, and said the model is available now to Canva users and to developers through its public API.

Canva said Craft debuted as the fourth most advanced image generation model on DesignArena, a crowdsourced benchmark for AI design, and that the placement makes Canva Research the world's second most advanced lab for image generation. DesignArena's own image leaderboard, read on October 23, 2026, listed Canva Craft fourth with an Elo rating of 1,329, behind three OpenAI models: GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT Image 2. The lab-level claim is Canva's own reading of that table, which ranks models rather than labs.

Craft uses a two-stage architecture, Canva said in its announcement. A planning model first reads the prompt and writes a composition plan covering which elements appear, their spatial relationships, relative scale, visual hierarchy and where text goes. A separate rendering stage then produces the image at up to 2K resolution across more than 30 aspect ratios. Canva said prompts can be written in any language, and that the model can render text in 55 of them.

Most image models learn mainly from photographs. Canva said Craft was trained on millions of human-made templates from its own content library instead, spanning posters, social posts, presentations, infographics, menus and typography-heavy layouts. The company credits that design-first training for legible text and accurate placement of objects. Canva said Craft ranks first on DesignArena for People & Portraits, out of 87 models, and for Landscapes & Nature, out of 85, and second for Abstract & Patterns.

Cliff Obrecht, Canva's co-founder and chief operating officer, said in comments published by Capital Brief: "This is a huge moment for us, and hopefully it shows that you don't need to be in Silicon Valley to build world-class AI."

The release moves Canva from reselling other companies' image models toward training its own. TechCrunch reported that Canva has used a range of third-party models for generation and that an in-house model could cut its API costs. TechCrunch also reported that Craft images can be converted into layered designs with Canva Magic, the platform's image-to-design tool, that Canva plans third-party chatbot access through its Canva MCP server. Capital Brief reported that Canva brought the announcement forward after preliminary leaderboard results. Researchers from Leonardo.ai, the media generation platform Canva acquired in 2024, contributed to the model, TechCrunch reported.

Canva said it reduced Craft's inference steps through model distillation and built a custom inference engine for the planning stage, making it an order of magnitude faster than its first internal version. Those advances underpin Craft-fast and Craft-ultrafast, two coming variants the company said will generate images at a fraction of the cost of leading frontier models. Canva gave no price or date for either.

Canva said Craft is available to all Canva users globally through the Dream Lab tab and through its public API. Thurrott.com reported that using it in the app requires a paid subscription, while Geo.tv said free accounts can use it subject to AI usage limits. Canva listed the Canva MCP Server, which would let AI assistants generate images inside a larger task, as coming soon.

For people who make AI video, the practical effect is indirect. Images made in Canva feed its design templates and its video tools, and a design-first model tuned for legible text and clean layouts could cut the fixing-up that follows a generated frame. Whether that holds at scale is unproven, and Canva's quality and ranking claims remain its own until independent tests exist.

Canva has not said when Craft-fast, Craft-ultrafast or the Canva MCP Server integration will ship.