Canva Craft: What Canva’s New Image Model Means for Design

What is Canva Craft?

Canva Craft is Canva Research's first image foundation model built and trained from scratch, with a focus on design: the arrangement of objects, the placement of text, and the hierarchy that makes a visual easy to understand. Canva announced it on October 23, 2026, with access through Dream Lab and a public image-generation API.

Its defining idea is to plan a composition before rendering the image. That makes Craft particularly relevant to people creating posters, social graphics, presentation visuals, and other assets where attractive imagery is only part of the job.

For developers, there is an important qualification: the image-generation endpoint is currently a preview API. Public availability does not mean general availability or unrestricted use in a publicly distributed Canva app.

This article separates the model's announced capabilities from the access routes you can use today. For a broader shortlist of models to evaluate alongside it, browse the SeeAPI model catalog; this is not a Canva Craft integration page.

Research checked October 23, 2026. This is an analysis of official sources, not a hands-on benchmark. The cover is an editorial illustration, not an image generated by Craft.

Why planning before rendering matters

Canva describes a two-stage system. First, a planning model interprets the prompt and specifies the composition: elements, relative sizes, positions, visual emphasis, and text. A separate rendering model turns that plan into an image.

Consider a coffee-shop poster. A useful result needs more than an appealing cup: the shop name should be readable, the opening date should receive the right emphasis, and the background should leave enough room for the message. Craft's architecture explicitly addresses those relationships before producing the final pixels.

Canva says its training includes millions of human-made templates from its content library. It also describes fine-tuning and reinforcement learning informed by professional designers. That explains the model's design emphasis; it does not establish that every generated layout will be production-ready.

The practical question is how much correction a usable asset needs. A beautiful image with an incorrect date can still be a failed marketing deliverable. Review text accuracy, spacing, hierarchy, and export dimensions together.

Capabilities worth checking in your own workflow

According to Canva's launch announcement, Craft supports output up to 2K, more than 30 aspect ratios at the model/product level, and text generation in 55 languages. Prompts can be written in any language.

These capabilities suggest several useful evaluation tasks:

This is a proposed evaluation framework, not a report of tests we performed. Use a small set of representative briefs and record which corrections each output needs. That produces more useful evidence than selecting one impressive image.

Editable design requires a separate step

Canva says Craft's structure works well with Magic Layers, which can turn generated images into editable design elements. Keep that distinction clear: generating an image and converting it into editable elements are separate operations.

The image-generation API documents an image result and an optional saved Canva asset. It does not promise that the response is a complete, layered design document. A workflow that requires separately editable text, logos, or layout objects should validate the conversion step before committing to it.

Canva also identifies more advanced editing and reference-image capabilities as future work. Those should not be treated as already available features of the launch API.

How to access Canva Craft

Canva says Craft is available to Canva users globally through Dream Lab, accessible from the bottom-left panel in Canva. Availability and account allowance are different questions: check the allowance and any limits shown in your account before planning a batch of work.

The launch has several distinct delivery states:

The model's broad aspect-ratio support also needs a route-specific check. The API reference currently lists six accepted values: square, landscape, portrait, landscape_3_2, portrait_2_3, and portrait_4_5. Do not send an arbitrary custom ratio merely because the product announcement describes broader support.

Canva Craft API: the details that affect integration

The official image-generation reference identifies Craft with the model value craft_v1_2k. The operation creates an asynchronous job through POST /v1/image-generations; use the regional API base URL documented for your integration.

The following is an illustrative request body, not an executed test:

{
  "prompt": "Create a botanical exhibition poster. Place the exact title 'URBAN GARDENS' at the top, a single large leaf illustration in the center, and the text 'Saturday 10 AM' below. Use generous margins and a warm ivory background.",
  "model": "craft_v1_2k",
  "aspect_ratio": "portrait_2_3",
  "idempotency_key": "42ed67af-9acc-41c6-a362-64ce9a986b71"
}

Generate a new idempotency key for each new intended job; reuse the same key when retrying that job. The reference says keys persist for 24 hours and a repeated key returns the original job even if other request parameters differ.

An integration should account for the following:

Authorization: requests act on behalf of a user, require the asset:write scope, and require the user's image_generation capability.

Job lifecycle: poll the returned job until it succeeds or fails. Treat failure as a terminal outcome, not a reason to poll forever.

Model identity: pinning craft_v1_2k requests Craft specifically. If it cannot serve the request, the job fails; Canva does not silently substitute another model. Omitting the model allows automatic selection and fallback, so the output should not then be labeled as a Craft result without verification.

Result retention: download URLs last 24 hours. Save the result within that window, or use the documented asset-upload option when the workflow needs a Canva asset.

Credits: starting a job consumes the user's AI credit allowance. An exhausted allowance can produce credit_quota_exceeded; this is different from a malformed prompt.

The preview restriction matters before distribution: Canva's reference warns that breaking changes may occur without a new API version and that public apps using preview APIs cannot pass review for availability to all Canva users. Verify the current policy before moving an experiment into a customer-facing app.

What the launch ranking does—and does not—show

Canva reports that Craft debuted fourth among image-generation models on DesignArena. It separately describes Canva Research as the second-ranked image-generation lab. Those are different comparisons: a lab's position is not the same as an individual model's position.

Treat these as attributed launch-time results, not a permanent ranking or proof that Craft will outperform another model on your own briefs. We have not independently reproduced the evaluation. A practical comparison should use the same text, intended layout, target format, and review criteria across models, then count usable outputs and correction work.

Canva itself identifies fine print and technical drawings as areas it plans to improve. For assets containing dense labels, precise diagrams, or critical product details, include those requirements in the evaluation rather than relying on a general image-quality score.

Where Canva Craft fits

Craft is worth evaluating when composition and typography are central to the deliverable, especially if the next step is editing inside Canva. Its planning stage gives it a clear design-oriented proposition; the preview API gives developers a concrete route to investigate it, with constraints that need to remain visible.

Start with one representative brief, inspect the image at its actual delivery size, and test the editing handoff you need. For workflows that also require other image or video models, use SeeAPI's model market to review currently listed options and their documentation. Canva's own access routes above are the verified starting point for Craft; this article does not establish Craft availability on SeeAPI.