Canva Craft: Canva's First In-House Image Model Debuts 4th

Canva Craft is Canva's first in-house image model, and it landed near the top of the leaderboard on day one. The design company launched it on October 22, made it available to all users and to developers through its API, and says it debuted fourth on DesignArena's image generation rankings, behind three OpenAI models.

What actually happened

Canva announced Craft in its newsroom as the first foundation model "built and trained from scratch" by Canva Research, its in-house AI lab. The Verge reported the launch the same day and confirmed that Craft sits fourth on DesignArena, a crowdsourced benchmark for AI design output, behind OpenAI's GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT-Image-2.

Canva is leaning hard on the ranking. It says Craft's debut makes Canva Research the second most advanced AI lab for image generation on DesignArena, and that Craft placed first in two subcategories, People & Portraits and Landscapes & Nature, out of more than 80 models in each. It also claims second place in Abstract & Patterns and fourth in Products.

The model is live now. Users can find it in Dream Lab inside Canva, and developers can call it through Canva's Connect API, according to a post on Canva's developer forum.

How Canva Craft works

Most image models go straight from prompt to pixels. Craft splits the job in two. A planning model reads the prompt and writes a detailed plan of the composition: which elements appear, where they sit, how big they are, what the visual hierarchy is and what text needs to show up. A separate rendering model then turns that plan into an image.

Canva says this is why the model handles text and layout better than typical generators. It also feeds into Magic Layers, Canva's tool that turns a generated image into separate, editable design elements. Canva describes a JSON bounding-box approach that makes that conversion faster and more accurate.

Planning before painting is the whole pitch. If the text lands where you asked, you spend less time fixing it in an editor.

The training data is the other edge. Instead of mostly photos, Canva says it trained on millions of human-made templates, from posters and social posts to menus and infographics. After pre-training, the team used supervised fine-tuning and reinforcement learning with feedback from professional designers, and ran more than 25,000 head-to-head image comparisons against leading models. When automated scores and designers disagreed, Canva says it sided with the designer.

Speed, cost and what is next

Scale is a real constraint for Canva. The company says more than a quarter of a billion people use its product and create over 400 designs every second. To serve that, Canva says it used distillation to cut the number of inference steps and built a custom inference engine for the planning stage, making Craft an order of magnitude faster than its first internal version.

Two cheaper variants are coming, called Craft-fast and Craft-ultrafast. The Verge also reports that Canva plans to let AI agents plug into Craft through the open Model Context Protocol, which would put it inside agent workflows rather than only inside the Canva editor.

For context on how fast visual models are moving, our Elorian Glimpse visual model piece and Odyssey 3 world model story cover two other recent launches.

Why Canva Craft matters

This is a product company becoming a model company. Canva has used third-party models for years. Owning its own model gives it more control over cost, quality and features, and less dependence on the labs it now competes with on DesignArena.

Keep the caveats in mind. The rankings are crowdsourced and can move quickly as new models arrive, and Canva's subcategory claims come from Canva itself. Independent tests over the next few weeks will say more than launch-day charts.

What to do as a reader

Try it before you believe the hype. Canva Craft is in Dream Lab now, so test it on the thing you actually make, like a poster with lots of text, and compare it with the model you already use.

If you build tools, check the Canva Craft announcement and the Connect API docs for limits and pricing before you wire it into anything.

There is no Canva token. Any "Craft" coin riding this headline has nothing to do with Canva. Nothing here is investment advice.

Sources: Canva newsroom, The Verge on the launch and rankings and Canva's developer forum post on API availability.