What happened
- On October 22, 2026, Canva unveiled Canva Craft, its first image generation model, developed and trained from scratch by its in-house lab, Canva Research.
- It works in two stages. A planning model writes out the composition (elements, positions, scale, hierarchy and text) and a rendering model turns it into an image of up to 2K in more than 30 aspect ratios.
- According to the company, training drew on millions of human-made templates from its library, such as posters, presentations, menus and infographics, followed by supervised fine-tuning and reinforcement learning with feedback from its designers.
- It is available to all Canva users and to developers through the API. It renders text in 55 languages.
Why it matters
- Canva stops relying only on outside models to generate images. For teams that build assets with text, such as ads or posts, illegible typography was the weak spot of image generators, and that is what Canva says the planning stage addresses.
- Training on templates made by its community leaves a question the announcement does not answer: what consent the authors of those templates gave for this use.
- The ranking the company cites is DesignArena, a crowd-voted evaluation, not an independent audit.
The number
4th place. Craft debuted fourth among image models on DesignArena, according to Canva. In portraits and landscapes, the company reports first place out of 87 and 85 models.
Context
Canva says more than 250 million people use it and create over 400 designs per second. To serve that volume, it says it cut inference steps through distillation and wrote its own engine for the planning stage. According to TechCrunch, researchers from Leonardo.ai, which Canva bought in 2024, worked on the model.
What’s next
- Craft-fast and Craft-ultrafast, two cheaper variants, with no date announced.
- Advanced editing and reference images in future releases, according to the announcement.
Bottom line
Legible text inside an image is a frontier several teams are chasing: MOSS-Image tries to write paragraphs inside an image with 5 billion parameters. And Adobe opened 70 tools inside Slack, another way of taking design to where people already work.
Sources
- Canva Newsroom — Introducing Canva Craft (October 22, 2026)
- TechCrunch — Canva releases its first image generation model (October 22, 2026)
- Cover image: Tony Webster from Minneapolis, Minnesota, United States, CC BY 2.0, via Wikimedia Commons
Edited by Rodrigo Cornejo. How we select and verify: who writes these notes.
.jpg){kind=link}