Canva Craft

Canva’s first model of its own

Until October 2026, the image generation inside Canva AI ran on models from other companies. Craft is the first one Canva built itself. It was trained from scratch by Canva Research, a team that includes researchers from Leonardo.ai, which Canva bought in 2024. Canva co-founder Cliff Obrecht has called it the first of many models.

That makes Craft worth a page of its own, separate from Canva’s broader AI features. It is a foundation model with its own API, its own leaderboard position, and its own strengths and gaps, and developers can use it outside the Canva editor.

Trained on designs, not photos

The most distinctive thing about Craft is its training data. Most image models learn mainly from photographs. Canva says Craft was trained on human-made designs from its template library: posters, social posts, presentations, infographics, menus, and typography-heavy layouts.

Craft also works in two stages. The first reads your prompt and makes a plan: where elements go, how big they are relative to each other, what should draw the eye, and what the text says. The second renders the image from that plan. The goal is output that looks composed by a designer, with layout and hierarchy, rather than a striking picture with text bolted on.

Text is the practical test for any image model, and it is where Canva is making its strongest claims. You can write prompts in any language, and Canva says Craft renders text in the image in 55 languages. Output goes up to 2K resolution in more than 30 preset aspect ratios or a custom size.

Where you use it

In Canva: Craft sits in Dream Lab, Canva’s image generation space, selectable from the bottom-left panel. Canva says it is available to all users. You can make targeted edits to part of an image without regenerating the rest, and Canva’s Magic tools can convert a generated image into a layered design you can edit element by element. That last step is where Craft fits Canva’s workflow better than an outside model would.

Through the API: developers can call it through the Canva Connect API, using the craft_v1_2k model for image generation. Canva says MCP support is coming, which would let AI assistants such as ChatGPT and Gemini generate design-ready images through Canva.

How it ranks

Canva says Craft debuted fourth overall on DesignArena, a crowdsourced leaderboard where people vote between images in blind comparisons, with three OpenAI-family models ahead of it. It reports first place in the People and Portraits and Landscapes and Nature categories, second in Abstract and Patterns, and fourth in Products.

Crowdsourced preference rankings measure which image people like more in a side-by-side, not whether text is spelled correctly every time or whether output is consistent across a campaign. At least one outlet noted the ranking does not settle those questions. Test Craft on your own briefs, especially text-heavy ones, before relying on it for production work.

Pricing and commercial use

Canva had not published Craft-specific usage limits or API prices at launch, so how many images each plan allows is not yet documented. For commercial use of generated images, Canva’s general terms for AI-generated content apply; check them for your plan before using Craft output in paid work, as with any generator. Canva says two cheaper variants, Craft-fast and Craft-ultrafast, are in development.

What it is good at, and what it isn’t yet

Craft looks strongest for the work Canva users already do: social graphics, posters, event flyers, menus, and marketing images where layout and legible text matter. Generating directly into an editable Canva design is a real workflow advantage.

It is less proven for photorealistic product shots, where Canva reports a fourth-place category ranking, and for brand-consistent series, since reference-image support has not shipped. Advanced editing is also listed as future work. If you need fine control from reference images today, other tools are ahead.

Alternatives

• Ideogram, a long-standing choice for text inside images
• Midjourney, for stylized and artistic imagery
• Adobe Firefly, image generation inside Adobe’s tools with a commercial-use focus
• Krea AI, for fast, real-time image exploration
• Muse Image, Meta’s image model inside its apps

Related: Canva AI for the rest of Canva’s AI features, Midjourney vs Canva AI, and our text-to-image explainer.