Canva has released Canva Craft, the first image generation model built by its own research team. Until now the design platform has relied on models from other companies to power its AI image features. Canva Craft puts Canva in direct competition with image models from Google, OpenAI, ByteDance and Meta, according to TechCrunch's Ivan Mehta, who reported the launch on 22 October 2026.
For the millions of people who use Canva to make posters, social posts and slides, the more useful question is what Craft does differently and where it shows up. Here is what is known so far.
What Canva Craft does
Canva says Craft works in two stages. First, the model reads your prompt and makes a plan. That plan covers where elements sit relative to each other, how big they are compared with one another, the visual hierarchy, and any text that should appear. Second, it generates the image.
The output specs Canva has shared:
- Resolution: up to 2K.
- Aspect ratios: 30 options, sized for different platforms.
- Text: Canva says text in images comes out legible, a long-standing weak point for image models.
- Edits: Canva says you can make precise changes to one part of an image without the rest shifting.
The planning step is aimed at a familiar problem. General-purpose image models are good at a single striking picture, but they often struggle with layouts: a headline that misspells itself, a logo in the wrong corner, or a product shot squeezed out of frame. A model that plans spacing and text before it draws should, in theory, handle a menu or an event flyer better than one trained mainly to make photographs.
Trained on templates, not photos
Canva says it trained Craft on human-designed templates from its own library rather than on photographs. That library includes posters, social media posts, presentations, infographics, menus and typography-heavy layouts. It is a different starting point from most image models, and it explains Canva's emphasis on layout and text rather than photorealism.
Canva has not published benchmark results, so its claims about accuracy and text quality cannot yet be checked against rival models.
From flat image to editable design
The most practical feature for regular Canva users may be what happens after generation. Images made with Craft can be turned into layered designs using Canva Magic, the company's image-to-design tool. In plain terms, a generated poster can be split into separate pieces, such as background, text and shapes, which you can then move and edit in the usual Canva editor.
That matters because a flat AI image is hard to fix. If the date on a flyer is wrong, you normally have to regenerate the whole thing and hope the rest stays the same. A layered file lets you fix the date and keep everything else.
Where to find Canva Craft
- In Canva: Craft is available to all users through the Dream Lab tab.
- Through the Canva API: developers can call the model from their own apps.
- In chatbots, later: Canva plans to support image generation inside third-party assistants such as ChatGPT and Gemini through its MCP connection. It has not given a date.
Canva also says two cheaper versions, Craftfast and Craft-ultrafast, are coming "soon." Neither consumer pricing nor API pricing was disclosed at launch, and Canva has not said whether Dream Lab usage limits differ by plan. Readers in Africa should check Dream Lab in their own accounts, since Canva has not stated regional availability beyond "all users."
Why Canva built its own image model
There are two clear business reasons. The first is cost. Canva already runs several third-party image models, and TechCrunch notes that an in-house model could cut what Canva pays in API fees. Canva also says Craft gives it more control over quality and speed.
The second is people. Canva bought the media generation platform Leonardo.ai in 2024, and the company says Leonardo researchers contributed to Craft. Building on that team makes Canva less dependent on suppliers who are also, increasingly, its competitors. OpenAI and Google both now offer design and image tools inside their own chat products.
The launch also comes at a sensitive time for the company. TechCrunch reports that Canva cut its 2026 growth forecast and that outside analysts reportedly value it at $35 billion, down from $42 billion. A model that lowers costs and gives users a reason to stay inside Canva rather than generate images elsewhere helps on both fronts.
What this means if you design for a living
- Small businesses and social media managers are the obvious winners if text rendering is as good as claimed. Flyers, price lists and promo posts are exactly the formats Craft was trained on.
- Freelance designers should treat Craft as a first-draft tool. The layered output makes it easier to take a generated layout and refine it by hand.
- Developers building content tools can test the API now, but should budget carefully until Canva publishes prices.
Image models are becoming a crowded field, and we have tracked several smaller, cheaper launches this month, including the Recursive Flint coding model. Canva's advantage is not the model alone but where it sits: inside an editor that a very large number of people already open every day.
Open questions
- How much does Craft cost through the API, and will Dream Lab limits change?
- How does it compare with rival models on independent tests of text and layout?
- When will the ChatGPT and Gemini integrations go live?
- What rights do users have over Craft outputs for commercial use?
We will update this explainer when Canva publishes pricing and the faster Craft versions.
