
On October 22, 2026, Canva released Canva Craft, its first in-house image generation model. Until now, Canva relied on third-party models; Craft was trained from scratch by Canva Research, the company’s internal lab. It has more than 160 scientists and engineers, including the Leonardo.ai team Canva acquired in 2024.
How Canva Craft works
Craft generates an image in two stages. First, a separate model plans the composition: where each element goes, how big it is, what the focal point is and where text will sit. Then a second model renders the image from that plan. The approach mirrors how designers work: layout first, then the final artwork.
The key difference is the training data. Craft wasn’t trained on photos but on millions of human-made templates from Canva’s library: posters, social posts, presentations, infographics, menus and typographic layouts. The model was then fine-tuned with reinforcement learning on ratings from Canva’s professional designers, and its testing included more than 25,000 pairwise comparisons against leading models.
What it can do
- Up to 2K and 30+ aspect ratios — for any platform, from stories to banners.
- Legible text and precise edits. Canva says the model renders objects accurately, writes readable text and edits part of an image without changing the rest.
- From image to editable design. Canva can split a generated image into layers with Magic Layers, so you can keep editing it like a regular design.
How Craft ranks
For now, the rankings come only from Canva. At launch, Craft placed 4th among image models on DesignArena, a leaderboard that specifically judges design, and 1st in the People & Portraits (out of 87 models) and Landscapes & Nature (out of 85) categories. It ranked 2nd in Abstract & Patterns and 4th in Products. As of October 24, Craft isn’t on the overall Arena leaderboard, so there’s no independent way to compare it with GPT Image or Nano Banana 2.1 yet.
Availability and what’s next
Craft is already available to all Canva users worldwide in Dream Lab. Developers can connect it through Canva’s public API and the Canva MCP server, which also lets the model work inside chatbots like ChatGPT and Gemini. Canva is also preparing two cheaper, faster versions: Craft-fast and Craft-ultrafast.
How to make design images in AI2Face
For posters, banners and images with text, the AI2Face generator offers models with a proven track record on independent leaderboards: GPT Image, #1 on Arena; Ideogram for typography and bounding-box layouts; and Recraft for vector logos and icons. Run one prompt on several models and keep the best result. The cost is shown before you generate. For more on the task, see our guide to AI banner generators.
FAQ
What is Canva Craft?
Canva's first in-house image generation model, released on October 22, 2026. It was trained from scratch by Canva Research, a lab of more than 160 scientists and engineers that includes the Leonardo.ai team.
Where can I try Canva Craft?
In Canva, under Dream Lab — the model is available to all users worldwide. Developers can use it through Canva's public API and the Canva MCP server.
How is Canva Craft different from other models?
It was trained on millions of human-made templates rather than photos, and it works in two stages: a layout plan first, then the image. That's why Craft is strong at posters, menus, infographics and social media cards.
How does Canva Craft rank?
According to Canva, it debuted at #4 on the DesignArena leaderboard and #1 in the People & Portraits and Landscapes categories. As of October 24, 2026, Craft isn't on the Arena leaderboard yet.
Sources
3
- https://www.canva.com/newsroom/news/canva-craft/
- https://www.canva.com/newsroom/news/introducing-canva-research/
- https://techcrunch.com/2026/10/22/canva-releases-its-first-image-generation-model/












