Canva Craft: Image Generation, API Access, Pricing and Limits

Canva released Craft on October 22, its first foundation image model built from scratch, with global access through Dream Lab and a public API. It generates images at up to 2K resolution using a separate planning stage. The important developer caveat is that the generation API remains a preview.

What Canva Craft actually generates

According to Canva's announcement, Craft first plans the composition, including object positions, scale, hierarchy and text. A rendering model then turns that plan into an image. Canva says it supports more than 30 aspect ratios and text in 55 languages.

The model was trained with a substantial emphasis on human-made design templates. Its outputs can be processed by Magic Layers into editable elements. Craft-fast and Craft-ultrafast are upcoming variants, while Craft support through Canva MCP is still described as coming soon.

That sequence matters for a fairly ordinary task: making a launch graphic with a product image, a headline and a short explanation. Getting each object right is only part of the job. The headline needs space, the product needs the right emphasis, and the reader needs to understand what to look at first.

Planning those relationships explicitly is a useful design choice. It does not guarantee that every result will respect the brief, or that the resulting layers will need no cleanup.

Canva Craft API access and restrictions

The generation API overview lists CRAFT-1 2K as available with the model identifier craft_v1_2k.

Public API access does not mean the API is production-stable. Canva labels these generation APIs as preview, warns that breaking changes may arrive without a new API version, and says public apps using preview APIs cannot pass review for availability to all Canva users.

That is a material distinction for a developer deciding whether to put Craft inside a customer-facing product. A prototype and a distributable Canva app have different requirements. Check the applicable review path before committing to a launch date.

The create-image endpoint uses an asynchronous job:

API detail	Current documentation
Request	POST to /rest/v1/image-generations
Required fields	A prompt and an idempotency key
Authentication	User access token with asset:write
Required user capability	image_generation
Creation rate limit	20 requests per minute per user of the app
Prompt length	Up to 10,000 characters

Pinning craft_v1_2k gives you a specific model rather than automatic selection. If that model is unavailable, Canva fails the job instead of silently substituting another model. Omitting the model allows automatic selection and fallback.

For a comparison test, pinning is useful because you know what you are evaluating. For a workflow that values completing the job above a specific visual style, automatic selection may be more convenient. Those are different product decisions.

Canva Craft pricing and usage limits

The API consumes the user's AI allowance. If that allowance is exhausted, the endpoint returns a credit-quota error. A successful image's download URL is valid for 24 hours, so a workflow needs to retrieve it or save the result rather than treating that URL as permanent storage.

Canva's AI usage guide explains that shared monthly allowances vary by plan. It does not provide a Craft-specific dollar price per API image in the material reviewed for this article. Availability to all users should therefore not be read as unlimited free generation.

Without a verified per-image rate, a direct cost comparison with another image API would be misleading. Start by checking which allowance the intended account receives, how much a representative job consumes and how frequently the workflow will retry unsatisfactory outputs.

How to read the benchmark claim

Canva reports that Craft debuted fourth on DesignArena's image-generation ranking. That is a preference result, not proof that it will be the best model for every brand, language or revision workflow. Canva also identifies fine print and technical drawings as areas it wants to improve.

For a design team, the useful comparison is the work left after generation. Can the headline be changed without rebuilding the composition? Does the output remain usable when the same campaign needs another format? How much manual correction survives the move into editable layers?

Craft's most interesting opportunity is to reduce that unfinished work inside a tool people already use. The practical test is how quickly a generated image becomes something a team can confidently publish.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help AI founders turn technical launches and product updates into clear stories developers and founders can understand and share.

Book a 30-minute call.