Canva releases Canva Craft, a two-stage AI image generation model
Canva has released Canva Craft, an AI image model that uses a two-stage process to plan and generate images up to 2K resolution for all global users.
2 min read
What happened
Canva has released its first image generation model, named Canva Craft [1, 2]. Developed by its research team, including contributors from Leonardo.ai, which Canva acquired in 2024, the model uses a two-stage architecture [2].
In the first stage, the model interprets user prompts to create a plan involving visual hierarchy, relative scale, spatial relationships of elements, and information for text generation [2]. In the second stage, it generates the final image [2]. According to Canva, this method allows for greater precision, consistency, and control, specifically for complex images requiring legible text and precise object placement [1].
Key technical specifications and details include:
- The model can generate images at up to 2K resolution [1, 2].
- It supports more than 30 different aspect ratios [1, 2].
- It was trained on millions of human-made templates from Canva’s library, such as menus, presentations, infographics, social media posts, and typography-rich layouts [1, 2].
- The model is currently ranked fourth on the Design Arena leaderboard, following OpenAI’s GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT-Image-2 [1].
- Canva Craft is available now to all users globally via the Dream Lab tab, found under the three-dot option on the left-hand menu [1, 2].
- The model will also be available through the Canva API [2].
Canva plans to expand this technology by allowing AI agents to connect to Canva Craft through the open Model Context Protocol (MCP) [1]. This will allow support for image generation in third-party chatbots like Gemini and ChatGPT [2]. Additionally, the company plans to release two more efficient versions called Craft-fast and Craft-ultrafast in the future [1, 2].
Canva stated that the model includes reporting tools, privacy and safety controls, and robust input and output moderation [1]. Users can also use Canva Magic, an image-to-design transformer, to convert generated images into layered designs [2].
Who is affected
Canva Craft is available to all users globally [1]. The model is intended to compete with image generation models from Google, Microsoft, Meta, OpenAI, and ByteDance [1, 2]. The company’s valuation reportedly fell from $42 billion to $35 billion following a cut to its 2026 growth forecast, though it was on track for $4 billion in revenue last year [2].
Sources
- 1. Canva’s first AI image model plans designs before generating them — The Verge
- 2. Canva releases its first image generation model — TechCrunch
Drafted with AI assistance from the reports above and checked automatically for sourcing. Always confirm dates, fees and requirements on the official website before acting. Report an error · How we correct mistakes
Comments
No comments yet. Share a question or what you know.