
Canva Craft: Wait for Independent Tests Before Switching
Canva's first in-house image model ranks fourth on DesignArena, a vendor-cited result. Test it on your own prompts before moving any workflow.
Wait. Canva Craft, the first image model Canva Research trained in-house, ranks fourth on DesignArena by Canva's own account, and first in portraits and landscapes. That rank is the one number that matters, and no independent test has confirmed it. Compare Craft against your current model through the Canva API on your own prompts.
The Weights Desk · 3 min read- The bottom line is Wait: Canva reports a fourth-place DesignArena ranking, and we found no independent replication of it.
- Canva says the model was trained mainly on millions of its own design templates rather than photographs, which points to layout and text work as the strength.
- Canva reports the model is available to all users in Dream Lab and through a public API. Its MCP server is listed as coming soon.
- Canva has published no pricing, parameter count or architecture detail, so cost and reproducibility claims cannot be checked yet.
- The cheaper Craft-fast and Craft-ultrafast variants are announced but not yet released.
Canva has released Canva Craft, its first image generation model built in-house, and the verdict for teams that ship images at scale is to wait. Canva says Craft ranks fourth overall on DesignArena. Both The Decoder and Canva's own announcement trace that figure to the company, and we found no independent test of it. It is a promising lead, not yet a finding.
The one number: fourth on DesignArena
Canva's announcement says Craft debuted as the fourth most advanced image model on DesignArena. It also reports first place in People & Portraits out of 87 models and in Landscapes & Nature out of 85. It lists second in Abstract & Patterns and fourth in Products. These are Canva's own citations of a ranking, not results from a harness a reader can rerun.
What Canva says it built
Canva Research trained the model, according to the announcement. Canva says training drew on millions of human-made templates from its content library, with more emphasis on design than on photographs. Fine-tuning followed, using supervised learning and reinforcement learning with professional designer feedback. Canva also cites internal head-to-head comparisons covering more than 25,000 image pairs, but it has not published the comparison data.
Specs and access
Craft outputs up to 2K resolution across more than 30 aspect ratios and can render text in 55 languages, Canva says. It is available to all Canva users in Dream Lab and through the public API. The Canva MCP Server integration is listed as coming soon. Craft-fast and Craft-ultrafast, two cheaper variants, are announced but not yet released.
What is missing
No price, parameter count, architecture detail or published evaluation protocol accompanies the launch. The Decoder reports only that the model plans composition and text before generating, with no technical specifics. Claims about cost savings and quality control come from the company. Text rendering across 55 languages is a testable claim, but no third-party results are cited.
How to test it before you switch
Canva points developers to the public API as the quickest way to compare Craft with other models on their own prompts. Do that with a fixed prompt set drawn from your real workload, covering embedded text, non-Latin scripts and your required aspect ratios. Score blind against your current model. Until that comparison or an independent replication exists, treat the fourth-place rank as a lead.
The bottom line
The bottom line: Wait. The badge is Wait, not Use it and not Skip it, and the deciding number is fourth place on DesignArena, reported by the vendor. That is enough reason to run a trial and not enough to migrate. Template-heavy training suggests Craft may suit layout and text work better than photorealism, but that is inference, not a result. Revisit once the cheaper variants ship, prices are published and someone outside Canva reproduces the ranking.
- Should teams adopt Canva Craft now?
- Wait. The only ranking evidence is Canva's own citation of DesignArena, where it reports a fourth-place overall finish. Run the public API against your own prompts before switching any production workflow.
- What data was Canva Craft trained on?
- Canva says training used millions of human-made templates from its content library, with more emphasis on design than on photographs. It then applied supervised fine-tuning and reinforcement learning with feedback from professional designers.
- Where can developers use Canva Craft?
- Canva says Craft is available to all users in Dream Lab and through its public API. The Canva MCP Server is listed as coming soon.
- Canva launches Craft, its first in-house image model — The Decoder
- Canva Craft — Canva Newsroom