Photo: Steve A Johnson / Pexels
The design platform's research team built a two-stage model aimed squarely at rivals from Google, OpenAI, ByteDance, and Meta
Canva has spent years helping people make slick graphics without touching Photoshop. Now it wants to make the pictures too.
The design platform has released Canva Craft, its first image generation model. The model was developed by Canva’s own research team, and it enters a category already crowded with offerings from Google, OpenAI, ByteDance, and Meta.
How Canva Craft works #
Canva says the model operates in two stages. The first is less about pixels and more about planning.
When a user types a prompt, Canva Craft first interprets the request. It then builds a plan before generating anything visual.
That plan covers several design fundamentals. According to the company, it maps out spatial relationships between elements, their relative scale, and the visual hierarchy of the composition.
The plan also includes information used for generating text inside the image.
The road to Canva Craft #
Canva Craft does not arrive in a vacuum. The company has been steadily building out its AI stack around what it calls the Canva Design Model.
That model was built to produce editable, layered outputs rather than flat images. It powers features such as Magic Layers and Canva AI 2.0.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The Canva AI 2.0 preview, which features the Design Model with added capabilities, was released in April 2026. In June 2026, Magic Layers was integrated with Google Gemini and OpenAI’s ChatGPT, letting users turn AI-generated images into editable designs inside Canva.
Canva has also expanded the models it offers, adding Proteus and Lucid Origin. Image-generation endpoints for its Connect API became available in preview on October 2, 2026, opening the door for developers and enterprise customers to tap those tools directly.
Then, on October 22, 2026, Canva acquired Australian startup MagicBrief. The startup focuses on AI-driven creative performance analytics and has analyzed over $6 billion in ad spend.
Before that deal, Canva reported 240 million monthly users and approximately $3 billion in annual revenue.
What this means for Canva and its rivals #
The biggest shift here is strategic. Until now, much of Canva’s AI story involved integrating or partnering with outside models and building tools around them.
A proprietary image model changes that equation. Owning the model gives Canva more control over quality, cost, and how tightly generation ties into its editing tools.
It also reduces dependence on companies that are, increasingly, competitors. Google, OpenAI, ByteDance, and Meta all have the resources to build design features of their own.
The planning-first design of Canva Craft hints at where the company thinks it can win. General-purpose image models chase photorealism and creative range. Canva appears focused on structure: layout, scale, hierarchy, and legible text.
The MagicBrief acquisition fits that picture. Pairing in-house generation with analytics on ad performance points toward a pipeline where Canva helps create the creative and then helps measure whether it worked.
For enterprise customers, the Connect API is the piece to watch. If Canva Craft eventually becomes available through those endpoints, businesses could plug Canva’s model into their own workflows rather than going through the consumer app. Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our