Canva Built Its Own Image Model and It Debuts Fourth. Canva's Own API Docs Point Text-Heavy Prompts at a Different Model, and There Is No Per-Image Price.
Canva Craft launched on October 22, the first image model Canva Research trained from scratch: a two-stage plan-then-render design, up to 2K output, in-image text in 55 languages, free to try in Dream Lab on every tier. DesignArena puts it around 1,341 Elo, fourth overall behind three OpenAI models, while Canva's newsroom calls itself the world's second most advanced AI lab for image generation. The API is a preview with six aspect ratios, no published per-image price, and a model reference that recommends z_image_turbo when the image includes text. We read the newsroom post, the developer forum, and the leaderboard capture.
By T.W. Ghost
Canva launched Craft today, October 22, 2026, and it is a bigger step than it sounds: the first foundation image model Canva Research trained from scratch, rather than a licensed model behind a Canva interface. It is a two-stage design, plan then render, outputs up to 2K, handles in-image text in 55 languages, and is free to try in Dream Lab on every Canva tier. The API model ID is craft_v1_2k, in preview.
The newsroom post describes Canva as "the world's second most advanced AI lab for image generation," citing DesignArena, and in the same breath says Craft debuted as the fourth most advanced model. Both can be true, because one ranks labs and the other ranks models, and a reader should know which one they are being told. We read the newsroom post, the Canva developer forum announcement, OrcaRouter's capture of the DesignArena board, and the API notes. Here is what held up.
What Shipped
| Canva Craft | |
|---|---|
| Launched | October 22, 2026 |
| Built by | Canva Research, trained from scratch |
| Design | Two-stage plan-then-render; up to 2K; in-image text in 55 languages |
| Where | Dream Lab, free to try on every tier; API as craft_v1_2k in preview |
| API limits | Six aspect ratios (the editor offers 30+); jobs about 50 to 60 seconds per the developer forum; preview may break; public apps using it cannot pass Canva's app review |
| Billing | End-user AI credits; no published per-image price |
| DesignArena | About 1,341 Elo, fourth, behind GPT-Image-2.5 Sunburst (about 1,406), Flare (about 1,382), GPT Image 2 (about 1,362), per OrcaRouter's capture |
| Coming | Craft-fast and ultrafast variants; an MCP server (Canva) |
| AA image arena | No entry found |
Fourth model, second lab. DesignArena's model board has three OpenAI image models above Craft. Canva's "second most advanced lab" is a lab-level ranking on the same site. Print the one that answers the question you are asking; for a buyer, that is the model rank.
Canva's own reference sends text elsewhere. OrcaRouter, quoting Canva's model reference, reports that the docs recommend z_image_turbo when the image includes text, or for food, drink, and general prompts. Craft's headline feature is text in 55 languages; Canva's API guidance for text-heavy images names a different model. Confirm the line in the Canva API reference before you build on it, and then decide which claim you believe.
The preview economics are opaque. Billing runs through end-user AI credits with no per-image dollar figure published anywhere we could find. Do not let anyone quote you one.
Verified vs Unconfirmed: The Scorecard
| Claim | Verdict |
|---|---|
| Launched Oct 22; Canva Research; two-stage; up to 2K; 55 languages; Dream Lab on all tiers | Verified (Canva newsroom) |
| "World's second most advanced AI lab" and "fourth most advanced model" | Verified as Canva's wording (newsroom), citing DesignArena |
| DesignArena about 1,341 Elo, fourth, behind three OpenAI models | Secondary (OrcaRouter's capture); re-pull the board before citing |
API craft_v1_2k in preview; six aspect ratios; credits billing; no per-image price | Verified (developer forum) and Secondary (OrcaRouter) |
Docs recommend z_image_turbo for text-heavy images | Secondary (OrcaRouter quoting Canva's reference); confirm on the reference |
| Subcategory wins (People and Portraits, Landscapes, and others) | Vendor (newsroom) |
| MCP, Craft-fast, ultrafast | Vendor, "coming soon" |
| AA image arena entry | Unverified (none found) |
Who Should Use It
Try Craft in Dream Lab if you already design in Canva; it is free on every tier and the two-stage design is Canva's own work, which is a first.
Hold off on the API for production: it is a preview that can break, public apps cannot pass review on it, the aspect ratios are a fraction of the editor's, and there is no price you can put in a budget. And if your images carry text, read Canva's own model reference before assuming Craft is the model Canva wants you to use for them.
The frontier is elsewhere: on DesignArena's model board the top three are OpenAI's, and Craft is a strong fourth on a capture we could not re-pull today. Our model comparison has the current standings on the language models; for images, re-pull the arena the day you decide.
Sources
- Canva newsroom: Canva Craft
- Canva developer community: Canva Craft is now available for image generation
- DesignArena leaderboard, image tab
- OrcaRouter: Canva Craft, DesignArena, and the API preview
*Not sure whether an in-app image model beats an API one for your design work? Take the free 2-minute quiz and get matched. Then read about the open image model that launched a week earlier with a price you can actually see: FLUX 3 Image Klein.*