Canva Craft AI Image Model: A Marketer's Guide to Access, Use Cases and Limits

Canva released Canva Craft on October 22, 2026, its first in-house image generation model. Craft plans a layout before rendering it, outputs up to 2K images across 30+ aspect ratios, renders legible text, and converts results into editable layers. It is in Dream Lab and the Canva API now. Canva has not published separate pricing, and early testers report invented copy and missing brand controls.

What is Canva Craft?

Canva Craft is an image generation model that Canva built and trained itself rather than licensing from OpenAI, Google or another lab. TechCrunch reported that Canva already runs several third-party image models, and that an in-house model will likely lower its API costs. Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to the work.

The design choice that matters most for marketers is the training data. According to Canva's announcement, Craft learned from millions of human-made templates (posters, social posts, presentations, infographics, menus and typography-heavy layouts) rather than mainly from photographs. Canva says designer feedback guided fine-tuning, and designers won when they disagreed with automated scores.

How does Craft's "plan, then generate" approach work?

Craft splits generation into two stages. A planning model first writes a composition plan covering which elements appear, where they sit, their relative scale, the visual hierarchy and the exact text. A separate rendering model then produces the image from that plan.

Canva says this produces three practical benefits:

• Legible text. Headlines, prices and labels come out readable, in 55 languages, per Canva.
• Localized edits. You can change one element without regenerating everything else.
• Editable output. Craft emits bounding-box data that Canva's Magic Layers tool uses to turn a flat image into a layered design you can adjust.

Canva's head of generative AI, JJ Fiasson, told SmartCompany the model delivers "exceptional results for jobs where the layouts and text need to be perfect."

How do you access Canva Craft?

You can reach Craft in two ways today, with a third on the way:

1. Dream Lab inside Canva. Open Dream Lab from the bottom-left panel or go to canva.com/dream-lab. Canva says Craft is available to all users globally. Thurrott, however, described it as available to paid subscribers, so check your plan before you build a workflow around it.
2. The Canva API. Developers can call Craft directly.
3. Canva MCP Server (coming soon). Canva plans to let assistants such as ChatGPT and Gemini generate Craft images through its MCP server. No date has been given.

Canva has not announced a separate price for Craft. One hands-on review by consultant Gaurav Tiwari reports that Craft draws on each plan's shared Canva AI allowance and that the API is in preview, capped at 20 requests per minute per user. Treat those details as one reviewer's findings until Canva documents them.

How should marketers use Craft day to day?

Use Craft for text-heavy, layout-driven assets, where it should have the biggest edge over photo-first models. Good first tests:

• Event posters and promo flyers with dates, offers and a call to action
• Menus, price lists and product comparison graphics
• Social carousels and story infographics in several aspect ratios

A simple workflow:

1. Write the prompt like a creative brief: say what goes where and give the exact copy in quotes.
2. Generate several variations and proofread every word. Don't assume the text is correct just because it is legible.
3. Use localized edits to fix single elements, then convert the image with Magic Layers to finish it in the editor.

How does Craft compare with GPT Image and Nano Banana?

On paper, Craft is near the top but not first. Canva says it debuted fourth on the crowdsourced DesignArena leaderboard. TechRadar notes that the three models ahead of it are from OpenAI, and that Craft ranked above several Gemini models, including Nano Banana 2.1. By category, Canva reports Craft ranked first for People & Portraits, first for Landscapes & Nature, second for Abstract & Patterns, and only fourth for Products.

Those rankings are an early snapshot built on public votes, and Canva's "second most advanced AI lab" claim compares labs, not individual models. For brands that sell physical products, the weaker Products ranking matters more than the overall rank.

What are the limits and criticisms?

The main risks are copy you didn't approve, brand controls that are still incomplete, and the usual unanswered questions about training data. In Tiwari's tests:

• Craft got the requested headlines and prices right, but added unrequested text in 3 of 4 poster variations, including invented opening hours.
• Single-element edits still caused small shifts across the image.
• Dream Lab's "Apply brand" option was unavailable, and only 6 aspect ratios were visible in his account.
• Magic Layers dropped some fine details, such as the dotted leader lines on a menu.

Canva itself says fine print and technical drawings still need work, and the cheaper Craft-fast and Craft-ultrafast variants have no release date or pricing. Tiwari also points out that Canva hasn't said how template creators were credited for the training data, and that its IP indemnity is aimed at eligible Enterprise customers. TechCrunch notes Canva recently cut its 2026 growth forecast, so cheaper AI serving matters to Canva too.

What should marketing teams do before spending on Craft creative?

Treat Craft as a fast way to make more options, not as proof that any of them will work. Cheaper generation means more variants, and that makes it more important to filter them before media money goes behind them.

A practical approach:

1. Proofread and brand-check every output. Invented copy is a compliance risk, especially in regulated categories.
2. Pre-test attention and clarity before launch. Tools like MarkGrid Neuro-Marketing estimate where viewers look and whether the key message lands, which is useful when an AI layout looks polished but buries the offer. Our explainer on the science behind AI-driven consumer insights covers how these predictions work.
3. Audit creator and partner assets the same way. If agencies or creators start submitting Craft-made visuals, Creator Audit can help check fit and consistency before you approve a partnership.
4. Shift spend toward proven variants. Use Budget Optimization to put budget behind the creative that tests best instead of spreading it evenly across untested options. If you're comparing tools for this step, see what to compare in creative asset testing platforms.

FAQ

Is Canva Craft free?

Canva says Craft is available to all users, but it hasn't announced standalone pricing. One reviewer reports that it draws on each plan's shared AI usage allowance, so free plans get limited use.

Where do I find Canva Craft?

In Canva's Dream Lab, or through the Canva API for developers.

Can Canva Craft generate readable text?

Canva says yes, in 55 languages. Early tests found accurate copy but some invented extra text, so proofread every output.

Can I edit Craft images after generating them?

Yes. You can make localized edits to individual elements, and Magic Layers can convert an output into an editable, layered Canva design.

Is Craft better than GPT Image or Nano Banana?

It depends on the job. Canva says Craft debuted fourth on DesignArena, behind three OpenAI models and ahead of several Gemini models. Test it on your own briefs before switching.

When will Craft work inside ChatGPT or Gemini?

Canva says support through its MCP server is coming soon but hasn't given a date.

Last updated October 25, 2026.