During this initial stage, Craft establishes the relative sizes and positions of objects, determines which elements should receive visual emphasis, and plans how written text will appear. It then uses that structure to generate an image at resolutions of up to 2K, supporting 30 aspect ratios for different formats and platforms.
Canva says this approach improves the accuracy of visual details and text rendering while allowing users to make targeted changes without unintentionally altering other parts of an image.
The model's training also differs from systems that rely primarily on photographic material. Canva used professionally created designs from its template library, including presentation slides, advertisements, social media graphics, posters, menus, infographics, and typography-focused layouts.
That training approach reflects Canva's emphasis on producing visuals intended for practical design work, where positioning, typography, and composition can be as important as the appearance of individual objects.
Craft builds on the company's existing investment in proprietary AI technology. Canva previously developed its Canva Design Model to generate editable designs with separate layers rather than delivering only flattened images. That technology supports products including Magic Layers and Canva AI 2.0.
The company introduced a preview of Canva AI 2.0 in April 2026, followed by integrations with ChatGPT and Google Gemini in June that allowed users to transform AI-generated pictures into editable Canva designs.
Images created with Craft can similarly be processed through Canva's Magic Layers technology, giving users the ability to work with individual design components after generation.
Canva has also continued expanding its AI offerings through external models, including Proteus and Lucid Origin. However, developing its own image-generation technology gives the company more direct control over model performance, output quality, and operating expenses.
The company has identified both speed and cost management as benefits of owning its underlying image model. Canva is also developing two more economical variants, Craftfast and Craft-ultrafast, which it plans to release soon.
The launch comes as Canva manages the expense of expanding its AI products. The company had been on track to reach $4 billion in revenue last year but subsequently reduced its 2026 growth forecast. External analysts reportedly estimated that its valuation declined from $42 billion to $35 billion.
Canva is extending Craft beyond its main design interface. The model will be accessible through the Canva API, allowing developers and business customers to incorporate image generation into their own applications. The company also plans to support image creation through third-party assistants, including ChatGPT and Gemini, using Canva's Model Context Protocol integration.
These plans follow the October 2 preview release of image-generation endpoints for Canva's Connect API, which expanded access to the platform's AI capabilities for developers.
Canva has also been investing in technology that connects creative production with performance measurement. On October 22, the company acquired Australian startup MagicBrief, which specializes in AI-powered advertising analytics and has evaluated more than $6 billion in advertising spending.
The acquisition adds another capability to Canva's creative software business, although the company has not announced a specific integration between MagicBrief and Craft.
With Craft now available in Dream Lab, Canva is moving beyond incorporating outside image models into its products and developing more of the technology itself. The new system emphasizes structured layouts, editable design workflows, and typography, while its planned API and third-party integrations will extend those capabilities beyond Canva's own interface.
This analysis is based on reporting from Crypto Briefing & TechCrunch.
Image courtesy of Canva.
This article was generated with AI assistance and reviewed for accuracy and quality.
