Canva Craft Launches: What the First Canva-Built Image Model Claims and What to Check
Canva trained its own image model on design templates instead of photos. The claims are specific, the pricing is missing and the API details matter most.
Jump to 11 sections
Canva Craft is Canva's first in-house image generation model, announced October 22, 2026. It plans a prompt layout first, then renders up to 2K in any of 30 aspect ratios, and Canva says it was trained on human-designed templates, not photographs. It is available in Canva's Dream Lab tab and API, but pricing has not been published.
Most image models are trained to make pictures. Canva says Craft was trained to make designs: posters, menus, social posts and layouts where text and spacing carry the message. That is a different target, and it is why the launch matters for anyone who builds on Canva. Model Drop covers the apps and tools built on top of new models, and this brief covers what Canva claims, what is missing and what developers and marketers should check first.
What Is Canva Craft?
Craft is Canva's first image model built by its own research team. According to TechCrunch, Canva positions it against image models from Google, OpenAI, ByteDance and Meta. It is available to all Canva users through the Dream Lab tab and through the Canva API.
The company has been moving toward its own models for some time. It acquired Leonardo.ai in 2024, and TechCrunch reports that Leonardo's researchers contributed to Craft. Until now, Canva has relied on several third-party image models, which complicates both cost and consistency.
A first-party model gives Canva control over price, speed and behavior. That is the business case, and it is the same logic that pushed other platform companies to build their own models over the past two years.
How Does Craft Generate an Image?
Craft works in two stages. In the first, it interprets the prompt and produces a plan covering spatial relationships, relative scale, visual hierarchy and the text content. In the second, it generates the image at up to 2K resolution in any of 30 aspect ratios.
The planning step is the interesting part. Many image models struggle when a prompt asks for specific text in specific places, because they generate all pixels at once. A plan-first approach is meant to place elements deliberately before drawing them. Canva says the model renders accurate object details and "legible text."
Those are claims, not measurements. Text rendering is also the area where image models have disappointed most often, so it deserves the most scrutiny. Short phrases usually survive. Long paragraphs, small type and unusual fonts are where errors tend to show up.
Why Does Training on Templates Matter?
Canva says it trained Craft on human-designed templates from its library, such as posters, social posts, presentations, infographics, menus and typography-heavy layouts, not on photographs. That choice shapes what the model is good at.
A model trained on layouts should understand composition: where a headline sits, how much space a logo needs, how hierarchy guides the eye. It is less likely to excel at photorealistic scenes, which is a different skill. Expect strengths in design-style output and caution on realistic photos.
It also raises a question that every buyer should ask: what rights apply to the output? Canva's templates were made by designers working with Canva. Check the license terms for commercial use, and whether outputs built from template-style training come with any restrictions or indemnity.
How Do Edits and Layered Designs Work?
Craft is paired with editing features. Canva says it supports precise edits that leave the rest of the image unchanged, which matters for iterating on a design without regenerating everything.
Images can also be converted into layered, editable designs with Canva Magic, the company's image-to-design tool. TechCrunch notes Canva's own design-understanding model was released in 2025. In practice that means a generated image can become a set of movable text and shape layers.
That conversion is the real product bet. A generated poster you cannot edit is a dead end. One you can turn into layers fits into how designers and marketers already work. How cleanly the conversion handles text and fine detail is a thing to test on your own files.
What Does the API Offer, and What Is Missing?
The announcement says Craft is available through the Canva API, and that Canva plans to support image generation in third-party chatbots such as ChatGPT and Gemini through its Model Context Protocol (MCP) connection. MCP is a standard way for AI apps to connect to outside tools.
The coverage gives no pricing. It does say Canva expects the model to lower its API costs, since it currently uses several image models. Lower costs for Canva do not automatically mean lower prices for you.
Before building on it, a developer would want to know several things:
- Per-image pricing and any free-tier limits.
- Rate limits and typical generation time at 2K.
- Supported formats and whether layered output is available through the API.
- Commercial usage terms and content policies.
- Which aspect ratios and sizes are supported on each plan.
None of that appears in the launch coverage. Treat any production plan as provisional until the documentation fills in.
How Does It Compare With Other Image Models?
There is no independent comparison yet. Canva says it competes on cost and customization against models from OpenAI, Google and Microsoft, as one daily AI news roundup framed it. Claims about quality and speed control come from the company.
For a baseline on how other image models are priced, see our comparison of AI image model prices, Nano Banana 2.1 versus FLUX 3 Image. That piece shows how much per-image cost can vary, and how quickly a cheap model becomes expensive at volume.
When Canva publishes Craft pricing, put it next to those numbers. Include the cost of retries, since a model that needs three tries per usable image costs three times its list price.
What Is Canva's Business Reason for Building Craft?
Canva is under some pressure. TechCrunch reports the company was on track for $4 billion in revenue last year but cut its 2026 growth forecast, and that outside analysts say its valuation reportedly fell from $42 billion to $35 billion.
A cheaper in-house model protects margins when millions of users generate images. It also reduces dependence on competitors who may also sell design tools. Owning the model lets Canva tune it for its own product.
That context explains the emphasis on cost. It also suggests the company has a strong incentive to push users toward Craft inside Canva, so watch whether other models stay available.
Who Gets the Most Value From Craft?
Small businesses and marketing teams are the obvious winners. They need menus, flyers, social posts and presentation slides in volume, and they rarely have a designer for every piece. A model that gets layout and text roughly right on the first try saves real time.
Agencies and professional designers may see it differently. For them, the question is control: can they steer the output precisely, and can they hand an editable file to a client? The layered conversion feature speaks to that, but it needs testing against professional standards.
Developers building products on Canva's API are a third group. They care about predictable pricing, stable behavior between model versions and clear terms. A first-party model can make all three better, or it can introduce new dependence on one vendor.
What Are Craftfast and Craft-ultrafast?
TechCrunch reports that two cost-efficient versions, Craftfast and Craft-ultrafast, are in development and will be released "soon." Canva has not said how they differ from Craft in quality or price.
The naming points to a familiar pattern. Labs ship a flagship model first and smaller, cheaper variants afterward, so that high-volume use cases can run at a lower cost per image. Our explainer on AI model point releases covers how to read these staggered launches.
For buyers, the practical advice is to wait for the full lineup before committing to a pricing model. The fast variants may suit thumbnails and social graphics, while the flagship suits print work and final assets.
How Should You Test Craft Before Adopting It?
Run a small, realistic trial. Use prompts from your actual work, not demo prompts, and judge the results against what you ship today.
A practical test plan looks like this.
- Write ten prompts from real projects, including at least three with specific text.
- Generate each at the aspect ratios you actually use.
- Check every piece of text for spelling and legibility at normal viewing size.
- Try one edit per image and see whether the rest of the image stays unchanged.
- Convert two images to layered designs and open them in the editor.
- Log the number of tries needed per usable result and the time per image.
Our guide to AI agent memory systems is a useful companion if you plan to wire a model like this into an agent that makes designs on request.
The Bottom Line on Canva Craft
Craft is a credible bet on a specific idea: design-trained models should beat photo-trained ones at layouts and text. The two-stage approach, 2K output and layered conversion are all sensible. What is missing is pricing, documentation and independent results.
At Model Drop, our editors will update this brief when Canva publishes pricing and when outside tests of text rendering arrive. Your next step is to run the ten-prompt trial above on a free account before you build anything on the API.
