Canva has released Canva Craft, the first image generation model the company built itself, and it is already open to all Canva users in the Dream Lab tab. Canva says Craft renders legible text and makes precise edits, and that it was trained on designs rather than photographs. The company claims it competes with models from Google, OpenAI, ByteDance and Meta.
What is Canva Craft?
Until now, Canva leaned on image models from other companies for its AI features. Craft changes that. Canva Research built it, with help from researchers who joined from Leonardo.ai, the startup Canva bought in 2024. TechCrunch says an in-house model should help Canva save on API costs, and The Decoder adds that it gives the company more control over quality and speed.
The training data is the interesting part. Canva says Craft learned mostly from human-made templates in its own library: posters, social posts, presentations, infographics, menus and layouts heavy on typography. That fits how most people use Canva. You rarely need a photorealistic picture. You need a flyer with a headline, a clear hierarchy and text in the right place.
How does it work?
Craft works in two steps. First it reads your prompt and draws up a plan: where things sit, how big each element is, what gets emphasis and what text has to appear. Then it generates the image, at up to 2K resolution and in about 30 aspect ratios built for different platforms.
According to The Decoder, Craft can render text in 55 languages. That matters if you make anything bilingual, since accents and long words are a classic weak spot for image generators. Test it with your real copy before you trust it.
Editing is the other pitch. Canva says the model can change one part of an image without disturbing the rest, and the result can be turned into a layered design with Canva’s image-to-design tool, which The Decoder calls Magic Layers.
Is it actually good?
The performance claims come from Canva. On the DesignArena ranking, the company says Craft placed fourth overall and first in portraits and landscapes. The sources include no independent testing, so read the ranking as a hint, not a verdict. There are also no head-to-head speed or quality comparisons against the generator you may already use.
How much does it cost, and when can you use it?
Craft is available now to all Canva users through Dream Lab and to developers through the Canva API. Neither source gives pricing, credits or plan limits. The quickest check is to open Dream Lab in your own account and see what your plan allows.
Canva also plans to bring image generation to chatbots such as ChatGPT and Gemini through MCP, a protocol that lets an AI assistant use third-party tools. No date was given. Two cheaper versions, Craft-fast and Craft-ultrafast, are expected soon, also without dates.
What it means in practice
Two things put this in context. The first is competition. ChatGPT and Gemini already make images, and anyone who just wants a standalone picture has plenty of choices, as we showed in our guide to making AI images in Gemini. Canva is betting on something different: the image is born inside a design editor, next to layers, social formats and the rest of your project.
The second is business. TechCrunch notes Canva was on track for about $4 billion in revenue last year but cut its 2026 growth forecast, and that outside analysts reportedly see its valuation dropping from $42 billion to $35 billion. Owning the model is one way to rely less on suppliers and set the product apart.
A caution: image generators that write text still slip, and a flyer with a misspelled word is worse than a flyer with no picture. Check every letter before you publish. If you want to understand why AI models can be so confidently wrong, our explainer on what a language model is covers the basics.
Is it worth trying now?
If you already live in Canva, yes. It costs a few minutes, and the Dream Lab tab is already there. Ask for something you usually make, such as a flyer with a title and three lines of text, and compare it with what your current generator produces. If the text comes out right and the targeted edit works, Craft earns a spot in your workflow. If not, you lost nothing. If you do not use Canva, there is no rush: the model only becomes useful outside the platform once chatbot support arrives.
