Canva Now Has Its Own AI Image Maker, and Every User Can Try It

Canva has released Canva Craft, its first image generation model, and made it available to every user of its platform through the Dream Lab tab. TechCrunch reported the launch on October 22, 2026.

Canva's own research team built the model. Canva Craft can produce images up to 2K resolution, and it supports 30 aspect ratios. An aspect ratio is the shape of an image, such as a square or a wide slide. Those shapes cover the formats Canva users switch between, from social media posts to presentation slides and printed layouts.

According to TechCrunch, Canva Craft does not turn a prompt directly into a picture. It first reads the request and makes a plan. That plan covers how objects sit relative to one another, how large each one should be, which element draws the eye first, and what any text should say. The image is then drawn from that plan. Text has long been a weak point for image models, so planning the words and layout in advance is a deliberate choice.

Canva trained the model on human-designed templates from its own library, not on photographs. The training material is made up of layouts, text placement, and design decisions made by professional designers. Most general image models learn from camera photos instead. Researchers from Leonardo.ai also contributed. Canva's 2024 newsroom post about Leonardo.ai said that company's models can generate realistic photos, character portraits, anime, and painted styles.

Everyone on Canva can use Canva Craft through Dream Lab. Canva also says it will be offered through the Canva API, which lets other software connect to the tool. Canva plans to bring Canva Craft into third-party chatbots, including ChatGPT and Gemini, through Canva's MCP (Model Context Protocol) server, a connection standard that lets outside AI assistants reach Canva. That last step is a plan, not a finished feature, and no date has been given.

Images made with Canva Craft can be turned into layered designs using Canva Magic, the set of features that links generated pictures to Canva's editor. In March 2026, Canva announced Magic Layers, which it said is built on the Canva Design Model, the company's foundation model. A foundation model is a large AI model that other tools are built on. Canva also said Magic Layers is the first to generate designs that are editable from the start.

Canva is also developing two cheaper versions, named Craftfast and Craft-ultrafast, for later release. Image generation always involves a trade-off between quality and the cost of running the model. The existence of a faster version suggests Canva expects heavy use across its users.

This launch is part of a wider push. In October 2025, Canva introduced its Creative Operating System and described the Canva Design Model as the world's first AI model trained to understand the full complexity of design. Bloomberg reported at the time that the model let users create and edit posters and short videos. In April 2026, Canva introduced Canva AI 2.0, an AI design platform that makes designs, documents, and websites. The company says it runs on the same foundation model. Bloomberg also reported in April 2026 that Canva has 265 million monthly users.

The broader context here is that this image model sits inside a design tool rather than on its own. A standalone generator is judged on picture quality and how closely it follows the prompt. A model that writes into an editable document has to fit the rest of the work: brand colors and fonts, text that stays editable, and images that can be changed without starting over. Canva's choices to train on templates, plan layouts first, and export layered files all point toward that second test.

In my view, the training decision deserves more attention than launch coverage usually gives it. Templates hold layout habits that professional designers have refined over decades of print and screen work. If the model learns from those habits, its default results should fit more easily into existing design systems. Whether that holds across every prompt and style will become clearer as people use the tool at scale.

In my view, the distribution plan is one to watch. Reaching ChatGPT and Gemini users through Canva's MCP could place design tools inside assistants many people already use for writing and research. Most Canva users are not professional designers, so that could move some design work into chat windows. The plan is still unfinished, and the details of how it will work have not been published.

For developers, the API is the most practical part of the launch. Once Canva Craft is available there, other companies' products can build image and layout features on the same model that powers Dream Lab. Teams already using Canva face a more immediate question: whether the 2K limit, the 30 aspect ratios, and the layered export path meet their production needs. They can now test that directly.