Canva Craft: Canva’s first image model, built for design

A two stage model that plans the layout and the text before it draws, now open to every Canva user.

On 22 October 2026 Canva released Craft, its first image generation model, built by its in house lab Canva Research. It is available to all Canva users in Dream Lab and to developers through the public API, and it plans the layout and text of an image before drawing it, which matters for posters, cards and social posts.

Key facts

Canva announced Canva Craft on 22 October 2026, its first image model trained from scratch by Canva Research.
It is available to all Canva users globally in Dream Lab, and to developers through Canva’s public API. Access through the Canva MCP server for AI assistants is listed as coming soon.
It creates images up to 2K resolution in more than 30 aspect ratios, or a custom one, and can write text in 55 languages.
It works in two stages: a planning model writes the composition and the text, then a rendering model draws the image.
Two cheaper and faster versions, Craft-fast and Craft-ultrafast, are announced but not yet released.

What happened

Canva, the design platform, unveiled Craft as its first in house image generation model. Until now Canva relied on image models from other companies; Craft was developed by its own research team, with help from researchers who joined through the Leonardo.ai acquisition, according to TechCrunch.

The main idea is the architecture. Most image models go straight from a prompt to pixels. Craft first runs a planning model that reads the prompt and writes a detailed plan: which elements appear, where they sit, how big they are compared to each other, what is most important and what text must appear. A second model then renders that plan into the final image. Canva says this gives more legible text, more precise placement and the ability to change one element without redrawing everything else.

Training also leans on design. Canva says the data includes millions of human made templates from its library: posters, social posts, presentations, infographics, menus and text heavy layouts, not only photographs. Images made with Craft can be turned into editable layers with Canva’s Magic Layers.

Canva reports that Craft debuted fourth on DesignArena, a crowdsourced ranking of AI design, and first in its People and Portraits and Landscapes and Nature categories. These are Canva’s own figures based on that ranking.

Why it matters

Text inside AI images has long been the weak spot: misspelled words on a flyer, a date that comes out garbled, a headline squeezed into a corner. Canva is betting that planning the layout first fixes much of that, and it is putting the model where many people already make posters, invitations and social graphics.

For everyday users this means two things. First, a design tool now generates the image and then lets you edit its parts, so a good first draft is worth more than ever. Second, prompts that describe structure get rewarded: the exact words, where they go, what the focus is, the format and what to leave out. A vague request like “a nice autumn poster” leaves the planner guessing.

For small businesses and creators, the public API also means Craft can appear inside other tools, such as ad builders, online shops and print services.

How to try it

Open Canva, go to Dream Lab in the bottom left panel and pick Craft. A local event card is a good first test, because it combines an illustration with text that must be spelled exactly.

Use our free prompt Greeting card for a local festival and fill the four fields: – The festival: name and date, for example “Chestnut festival in our village, second weekend of October”. – Who it is for: for example “Neighbours and families of the village”. – Exact message on the card: the words you want printed, for example “Join us on 10 and 11 October. Good food, music and friends.” – Illustration style: watercolour, flat illustration or paper collage.

The prompt already asks for a vertical 5:7 format with safe margins for printing, generic village scenery, no real people and no logos. It also asks the generator to check the spelling letter by letter and, if a word comes out wrong, to give you a second version with an empty banner so you can add the text yourself in the editor.

When the image arrives, check three things: every word and date matches your message, nothing important touches the edges, and the main element is clear at a glance. If something is off, change one instruction at a time instead of rewriting the whole prompt. Then use Magic Layers if you want to move or edit single elements.

Try it

Greeting card for a local festival

Frequently asked questions

Is Canva Craft free?

Canva says Craft is available to all Canva users globally from 22 October 2026, inside Dream Lab. The announcement does not give pricing details or usage limits, so check what your own Canva plan includes when you open Dream Lab.

Can I use Canva Craft from ChatGPT or Gemini?

Not yet. Canva lists access through its MCP server, which connects Craft to AI assistants and automated workflows, as coming soon. Today you can use Craft inside Canva or, if you are a developer, through Canva's public API.

Why is Craft better with text in images?

Canva says the planning stage decides the composition and the exact text before anything is drawn, and that the model was trained on many text heavy design templates. You still get the best results by writing the words in quotes and asking for them to be spelled exactly.

Sources