Canva says Craft is its first foundation model built from scratch

Canva says Craft separates planning an image’s composition from rendering it.[1] In its planning stage, the model specifies which elements should appear, how they relate spatially, their relative scale and visual hierarchy, and any text in the image.[1] The rendering stage turns that plan into an image at up to 2K resolution.[1] Canva says Craft can produce complex layouts with legible text and precise object placement, and can make targeted changes without unnecessarily altering the rest of the image.[1] This design may be relevant to teams that need control over layout as well as image content.

Craft supports more than 30 aspect ratios and lets users enter a custom ratio.[1] Canva says users can write prompts in any language, and the model can generate text in 55 languages.[1] Its outputs can be transformed into fully editable design elements using Magic Layers.[1] Canva says it trained Craft on millions of human-made templates from its content library, including posters, social posts, presentations, infographics and menus.[1] These capabilities place image generation alongside format selection and further editing within Canva’s design workflow.

Canva says its professional designers gave feedback during supervised fine-tuning and reinforcement learning, with the goal of improving instruction following, typography and visual quality.[1] The company says more than 25,000 image pairs were assessed in head-to-head comparisons with leading models.[1] Canva reports that Craft debuted as the fourth most advanced image-generation model on DesignArena, ranked first for People & Portraits among 87 models and first for Landscapes & Nature among 85 models, and placed second for Abstract & Patterns and fourth for Products.[1] These are company-reported results across named categories.

Canva says Craft is available globally to all Canva users, and developers can access it through the company’s public API.[1] Integration with AI assistants and automated workflows through Canva MCP is listed as coming soon.[1] Canva also says two variants, Craft-fast and Craft-ultrafast, are forthcoming and intended to provide high-quality image generation at a fraction of the cost of leading frontier models.[1] The company says model distillation reduced the required inference steps to a fraction of the original model’s, while a custom inference engine made Craft an order of magnitude faster than its first internal version.[1] TechCrunch reported that Canva developed Craft to compete with image models from Google, OpenAI, ByteDance and Meta, and that Leonardo researchers contributed to the model.[2]