Canva Craft: The AI Image Tool That Plans Layout Before It Draws
Canva announced Craft on October 22, 2026, and it is the company’s first state-of-the-art image-generation model, developed and trained from scratch by Canva Research. Craft generates images at 2K resolution in any aspect ratio, and it is available inside Canva and through Canva’s public API. What makes it different from other image models is the order in which it works. Craft plans the layout and the text first, and only then renders the image. That single choice is why its posters, social posts and presentations look designed rather than generated, and it is the idea this article explains in full.
Canva Craft is Canva’s first image-generation model built from scratch by Canva Research. It creates images at up to 2K resolution in more than 30 aspect ratios, plans layout and text first, and is available to all Canva users including free accounts through Dream Lab.
What Canva Craft is
Canva Craft is an image-generation model, which means it creates pictures from a text prompt. It was announced on October 22, 2026, and it is the first model of its kind that Canva has developed and trained from scratch. The training data is the interesting part. Craft was trained on millions of Canva’s own templates, which are human-designed posters, social posts, presentations and infographics from Canva’s library. It was not trained on photographs. That choice shapes what the model is good at. It understands how a poster is laid out, where a headline sits, and how text and images work together, because it learned from designs that people actually made and used.
The team behind Craft is Canva Research, a 160-person team that built the model from scratch in about six months. Six months is a short time to build a state-of-the-art image model, and it shows how much Canva has invested in its own AI lab. The lab’s job is to build models that understand design, not just models that understand pictures. Craft is the first result of that work, and it was followed a week later by Brush, the second in-house model, which edits designs rather than drawing them. The two models are the two halves of Canva’s AI plan. Craft makes the image, and Brush changes the design around it. Together, they are the reason Canva’s AI push in late October and early November 2026 matters.
Canva Craft in Dream Lab
Canva Craft lives in Dream Lab, which is Canva’s AI image tool, and it is available to all Canva users including free accounts. That is an important detail, because many AI image tools are locked behind a paid plan from day one. Craft is not. Every Canva user can try it, subject to each plan’s usage allowance. There is no separate price for Craft, and it spends the plan’s shared monthly AI allowance, which is the same pool that other AI features draw from. Free users get the allowance that comes with the free plan, and paid users get a larger one. The model itself is the same for everyone, and the difference is how much can be used each month.
Dream Lab is where the image is made, and the workflow is simple. A prompt describes what is wanted, and Craft produces the image. Because the model plans the layout and the text first, the result arrives with the structure already in place, which means less fixing afterwards. For a free user, this is a genuine chance to try a state-of-the-art image model without paying. For a Canva Pro user, the larger allowance means more images per month. The shared allowance is worth understanding before starting a big project, since heavy use of Craft will use up the same pool that other AI features need. For most people, the free allowance is enough to learn what the tool can do, and the paid allowance is there for heavier work.
Canva Craft output: 2K, aspect ratios and languages
The output of Canva Craft is where the numbers matter. Images come out at up to 2K resolution, which is sharp enough for social posts, presentations and printed posters at normal sizes. The aspect ratio is flexible, with more than 30 aspect ratios to choose from, so the same prompt can produce a tall story, a wide banner, a square post or a presentation slide without the image being cropped or stretched afterwards. That range matters because different platforms need different shapes, and a model that only makes one shape forces the user to crop and lose part of the image. Craft avoids that by making the right shape from the start.
| Detail | Canva Craft |
|---|---|
| Announced | October 22, 2026 |
| Built by | Canva Research |
| Resolution | Up to 2K |
| Aspect ratios | More than 30 |
| In-image text | 55 languages |
| Available to | All Canva users |
| Official site | Canva official site |
| Download the app | Download |
Language support is the other strength of the output. Prompts can be written in any language, which means the user does not need to translate an idea into English before the model understands it. In-image text, the words that appear inside the picture itself, can be rendered in 55 languages. That is a large number, and it matters for anyone making posters or social posts for an audience that does not read English. Many image models struggle with text inside pictures, producing garbled letters or misspelled words, and the problem gets worse in other languages. Craft’s design-first approach helps here, because the text is planned as part of the layout before the image is rendered, rather than being painted on afterwards. The result is text that is more likely to be readable and correctly spelled, in more languages than most models support.
Canva Craft Magic Layers
One of the most useful features of Canva Craft is Magic Layers, which converts the output into a layered, editable design. A normal AI image is a flat picture, and changing it means starting again or editing pixels by hand. With Magic Layers, the image Craft produces can be turned into layers that can be moved, resized and edited inside Canva. The headline can be changed, the background can be swapped, and the layout can be adjusted, all without regenerating the whole image. That is the feature that turns Craft from a picture generator into a design tool, and it is the reason the model fits so well inside Canva.
Magic Layers works because of how Craft was trained. Since the model learned from human-designed templates, it understands that a poster is made of parts, a headline, a background, an image, a block of text, and it keeps those parts separate in its own understanding of the design. When the output is converted through Magic Layers, those parts become real layers in the Canva editor. The practical benefit is time. A poster that needs a different headline does not need a new image. It needs one text edit. A social post that needs a different background colour does not need a new prompt. It needs one click. For anyone who makes many versions of the same design, for different platforms or different languages, that editability is worth more than the image itself. It is also the feature that connects Craft to Brush, the editing model that came a week later, since both models are built around the idea that a design should stay editable.
Canva Craft vs Canva Brush
Canva Craft and Canva Brush are the two in-house models Canva launched a week apart, and they do opposite jobs. Craft draws the image, and Brush edits the design. Craft was announced on October 22, 2026, and Brush followed on October 29, 2026. Craft is the first announced in-house model, and Brush is the second. Craft lives in Dream Lab and is available to all Canva users including free accounts. Brush is available to Canva Pro subscribers first, with wider availability coming soon. Craft creates something new from a prompt, and Brush changes something that already exists, while keeping every layer editable. The two models are designed to work together. A user can generate an image with Craft, turn it into a layered design with Magic Layers, and then edit it with Brush.
The third move in Canva’s AI push came on November 3, 2026, when Canva expanded its partnership with OpenAI, giving the platform broader access to GPT-6 Astra and GPT-5.6, according to Startup Daily. OpenAI already powers Canva features including Canva Code, which lets users create interactive designs, websites and apps by describing what they want. The OpenAI models sit alongside the in-house models rather than replacing them. Canva builds its own models where design understanding matters most, and it uses OpenAI models where a large general model fits better. Readers who follow AI tools will see the same pattern elsewhere. Google’s Dreambeans app connects several Google services into one AI assistant, and the Brahma AI guide covers an AI tool built for voice and lip-sync in films. Canva’s version of the idea is design. Craft for images, Brush for editing, and OpenAI models for the rest, all inside one product.
Canva Craft price and availability
Canva Craft has no separate price. It is available to all Canva users including free accounts, subject to each plan’s usage allowance, and it spends the plan’s shared monthly AI allowance. That means the cost of using Craft is already inside the Canva plan, and there is no extra charge for the model itself. Free users can try it within their allowance, and paid users get more usage each month. There is no launch date for general availability beyond the announcement, because the model is already available inside Canva through Dream Lab. What is still to come are the two cheaper variants, Craft-fast and Craft-ultrafast, which Canva has announced with no dates or prices. Those variants will trade some quality for speed and lower cost, but Canva has not said when they will arrive or what they will cost.
Developers can call Craft through Canva’s public API, using the model value craft_v1_2k. That is the technical name for the model in the API, and it is how outside apps and tools can generate images with Craft. Brush does not have a public API, which is one of the differences between the two models. The other public signal for Craft is its debut on DesignArena, a leaderboard where image models are ranked by users. Craft debuted in 4th place, with an Elo score of 1340, the day after launch. That is a leaderboard result, not a Canva claim, and it is worth reading as an early signal rather than a final verdict. A 4th place debut for a first model from a design company, on a leaderboard full of specialist image labs, is a strong start. It suggests the design-first training is working, and it gives users a reason to try the model for themselves rather than taking anyone’s word for it.
Who should try Canva Craft
Canva Craft suits anyone who makes posters, social posts, presentations or infographics and wants an AI image tool that understands layout. It is free to try for all Canva users, so there is no cost to testing it on a real project. The layout-first approach makes it a good fit for designs where text matters, such as event posters, sale announcements and social posts with a headline. The 55 languages for in-image text make it useful for audiences outside the English-speaking world, and the more than 30 aspect ratios mean the same idea can be made in every shape a platform needs. Magic Layers is the feature that makes it practical for repeated work, since a design can be edited instead of regenerated.
Canva Craft is a strong first image model because it is built for design rather than for pictures alone. It was trained on human-designed templates, it plans layout and text before rendering, it outputs up to 2K in more than 30 aspect ratios, it takes prompts in any language and renders in-image text in 55 languages, and it is available to every Canva user through Dream Lab. Magic Layers keeps the result editable, the public API lets developers use it, and the DesignArena debut in 4th place with an Elo of 1340 is an early sign that the approach works. The cheaper variants, Craft-fast and Craft-ultrafast, are announced but undated, so they are worth watching. For now, the model itself is ready, it is free to try, and it is the most design-aware image tool Canva has ever shipped. Open Dream Lab, write a prompt in any language, and see what a model trained on real designs can do.
Keep Reading