Canva launches Craft, its first in-house AI image model
Canva Craft plans a design before generating the image, focusing on posters and social graphics with legible text.
Canva, the design platform popular with non-designers, has launched its first in-house image generation model. It is called Canva Craft, announced on 22 October 2026. Unlike most image models, Craft was built specifically for design work such as posters, social posts and infographics.
1. What Canva launched
Canva Craft is an image generation model built by Canva's research team, with contributions from researchers at Leonardo.ai, which Canva acquired in 2024. It competes directly with image models from OpenAI, Google, ByteDance and Meta.
According to Canva, Craft was not trained mainly on photographs. Instead, it learned from millions of human-made design templates in Canva's library, including posters, social media posts, presentations, infographics, menus and typography-heavy layouts.
2. How it differs: plan first, generate later
Most image models go straight from prompt to pixels. Craft uses a two-stage process. The first stage reads your prompt and builds a plan: where objects sit, their relative scale, the visual hierarchy and details for any text. The second stage generates the image from that plan.
Canva says this produces images with accurate object details and legible text. Maximum resolution is 2K, with more than 30 aspect ratios for different platforms.
3. Edit after generating
Images generated with Craft can be converted into layered designs using Canva Magic. That means you can change one part of an image without touching the rest. Canva's Magic Layers feature already offered something similar before this.
For small business owners and content creators, that saves time. You do not have to regenerate an entire image just to change one word.
4. Where to try it
Craft is now available to all Canva users worldwide through the Dream Lab tab. It is also open through the Canva API for developers.
Canva also plans integrations with third-party chatbots such as ChatGPT and Gemini through MCP (Model Context Protocol). Two cheaper, faster versions, Craftfast and Craft-ultrafast, are coming later.
5. Where it stands in the market
On the Design Arena leaderboard, a crowdsourced benchmark for AI image models, Craft ranks fourth. According to The Verge, it sits behind three OpenAI models: GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT-Image-2. Canva itself claims it is now the world's second most advanced AI lab for image generation on the same leaderboard, so treat that as a company claim.
The move also helps Canva cut costs. Canva previously used image models from several third parties, so owning a model gives it control over cost, quality and speed.
What it means for you
If you make posters, ads or social posts for a business, there is now another option besides the usual image models. It is free to try if you already have a Canva account.
- Try it in Dream Lab: open the three-dot menu in Canva, select Dream Lab, and type your prompt.
- For text in images: Craft is designed so poster text stays legible, which suits written ads.
- Edit, don't regenerate: convert Craft output into a layered design and change just one element.
- Wait for the fast versions: if you generate images in bulk, Craftfast and Craft-ultrafast are on the way.
- Read rankings carefully: benchmark positions differ by source, so test it on your own use case.
Sources
- Canva releases its first image generation model (TechCrunch)
- Canva's first AI image model plans designs before generating them (The Verge)
- Canva enters the AI image generation race with Craft (Digital Trends)