Editorial illustration for Canva’s Craft, first AI image model, plans designs before generating them
Canva's Craft AI Plans Designs Before Generating
Canva built its own image generation model, and it works differently from the ones already on the market. The company announced Craft on Thursday, its first text-to-image system, built to compete with offerings from OpenAI, Microsoft, and Google. Rather than generating pixels straight from a prompt, Craft runs a two-stage process that plans a design first, then builds the image around that plan.
Canva says the approach came out of training on millions of templates pulled from its own content library, covering formats like social posts, presentations, infographics, and menus. That focus on design-heavy material shows up in what the company says Craft is built to handle: layouts with legible text, specific object placement, and the kind of visual structure that trips up most general-purpose image generators.
The launch fits a pattern Canva started earlier this year. In April, the company rolled out prompt-based editing that lets users describe changes to a design instead of manually adjusting it, along with layer support for AI-generated images so edits could be made to one part of a design without disturbing the rest. Craft extends that strategy into image generation itself.
Canva is diving even deeper into AI with the launch of an image generation model that competes directly with those from OpenAI, Microsoft, and Google. The model, called Canva Craft, uses a two-stage architecture that allows it to plan a design before generating an image, allowing for “greater precision, consistency and control,” according to an announcement on Thursday.
Why this matters
Canva building its own model instead of leaning on partners signals it wants to own the full design pipeline, not just license someone else's output. The two-stage planning architecture is the interesting part: generating a layout or structure before rendering pixels is a different bet than the one-shot diffusion approach most consumer tools still use, and it suggests Canva is optimizing for repeatable, editable output rather than one-off novelty images. For founders building on top of OpenAI, Midjourney, or Google's image APIs, this is a reminder that the platforms sitting closest to end users, the ones who own the distribution and the brand templates, can decide to stop being customers and start being competitors.
For researchers, the real test isn't the 2K resolution or the 30 aspect ratios Canva is touting. It's whether "plan then generate" actually produces fewer of the broken hands, warped text, and inconsistent branding that plague marketing-focused image tools today. Worth watching whether Craft stays Canva-exclusive or shows up as an API others can build on.
Related Reading
AI Tools & Apps Claude gains shared context in Excel, PowerPoint; Microsoft adds Copilot Cowork
AI Tools & Apps OpenAI Developer: AI Agent "Swarms" Waste Tokens, Create "Coordination Tax
AI Tools & Apps LG's recent webOS update adds Microsoft Copilot app, now removable
From Archives NVIDIA Minerva delivers dynamic, end-to-end observability for AI factories
From Archives Mistral and Nvidia Team Up to Speed Industrial Design With Physics AI
Common Questions Answered
How does Canva's Craft image generation model differ from competitors like OpenAI and Microsoft?
Canva Craft uses a two-stage architecture that plans a design first before generating the actual image, rather than creating pixels directly from a prompt like most competitor models. This approach allows for greater precision, consistency, and control in the generated output compared to the one-shot diffusion approach used by most consumer tools.
What training data did Canva use to develop its Craft image generation model?
Canva trained its Craft model on millions of templates pulled from its own content library. This extensive training on Canva's existing design templates informed the development of the two-stage planning process that distinguishes Craft from other image generation systems.
Why is Canva's two-stage planning architecture significant for design generation?
The two-stage planning architecture generates a layout or structure before rendering pixels, which optimizes for repeatable and editable output rather than one-off novelty images. This approach signals that Canva is prioritizing practical, production-ready designs that users can modify and refine, rather than static generated images.
What does Canva's decision to build its own image model instead of licensing competitors' models indicate?
By developing Craft internally rather than partnering with OpenAI, Microsoft, or Google, Canva signals its intention to own the full design pipeline and control its entire output. This strategic move demonstrates Canva's commitment to maintaining independence in AI capabilities rather than relying on third-party solutions.
Further Reading
- 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, and it's thinking beyond pretty pictures - Digital Trends
- Canva Releases Its First AI Model for Images - Thurrott
