Canva Launches Craft, Its First Public AI Image-Generation Model

Craft placed behind three OpenAI models on the leaderboard but ahead of several other GPT, Grok and Gemini models, including Nano Banana 2.1, according to TechRadar. Canva said the result makes Canva Research, its internal research division, the second-highest performing AI lab for image-generation models behind OpenAI. The company has tripled the size of its internal lab over the past year.

The model ranked first among 87 models for people and portraits, first among 85 for landscapes and nature, and second for abstract designs and patterns, TechRadar reported. Canva said Craft plans an image's composition before generating it, deciding where individual elements should appear, how the composition should be structured, where words should appear and how the result can remain cohesive and readable.

Craft was trained on millions of human-made templates in Canva's library, giving it an understanding of layout, visual hierarchy and typography that differs from models trained on flat images, according to the company. Canva said this training is suited to marketing materials and social media graphics.

Canva said developing in-house models gives it more control over the economics of generative AI than relying on third parties, as unpredictable costs for agentic AI affect enterprise users. The company is also offering Craft through an API so developers can integrate it into other platforms. Canva said future generations of Craft will be faster and more affordable.

Editor's Summary

Canva has entered the public AI image-generation market with Craft, a model trained on its template library and ranked fourth on DesignArena at launch. Its API availability and top scores in portraits, landscapes and abstract designs make it a direct offering from Canva rather than a third-party integration. Canva says in-house development gives it more control over generative AI costs and that future versions will be faster and cheaper.