Canva releases Craft, a design-first image generation model with Magic Layers Other

Canva announced Craft, its first in-house foundation model designed specifically for graphic design image generation, complete with layout planning and editable layers.

Trained on millions of design templates from Canva's library, Craft separates visual creation into composition planning and final rendering, allowing designers to manipulate individual scene components.

What it is

• Two-stage generation architecture. The model first maps the scene layout, calculating object relationships, proportions, and text placement before rendering pixels.
• Magic Layers separation. Internal segmentation breaks output images into editable vector objects and raster layers for direct manipulation in Canva's editor.
• High-resolution outputs. Craft renders images at resolutions up to 2K across arbitrary aspect ratios.
• Multilingual typography. The engine generates legible text in 55 languages while respecting composition and typographic balance.
• Developer and agent access. Craft is live in Canva's editor and public REST API, with planned support for a Canva MCP server to connect AI coding agents.

Sources

• Canva Newsroom: Introducing Canva Craft
• Canva Dream Lab
• Canva Developers