Robots Atlas>ROBOTS ATLAS
Canva Craft
AI Modelsโ€บCanva Craft

Canva Craft

Family: Canva Craft
Canva's first in-house image generation model. Two-stage (plan, then image up to 2K), trained on human-designed templates rather than photographs.
โœ“ Activeโœ“ Public accessImage generation๐Ÿ“ Canva Craft
Release date
22 October 2026
๐ŸขCanvaProducer
Access:APIHostedDeployment:โ˜ Cloud

Overview

Canva Craft is Canva's first in-house image generation model, developed by its research team. Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to its development.

The model works in two stages. It first interprets the prompt and builds a plan for the image โ€” spatial relationships between elements, their relative scale, visual hierarchy and text placement. In the second stage it generates an image at up to 2K resolution in up to 30 aspect ratios tailored to different platforms.

Rather than photographs, the model was trained on human-designed templates from Canva's library โ€” posters, social posts, presentations, infographics, menus and typography-rich layouts. As a result it renders legible text in images and allows precise, localized edits without changing the rest of the image.

Generated images can be converted into layered, editable designs via Canva Magic. The model is available to all users from the Dream Lab tab and through the Canva API; Canva also announced access from third-party chatbots (ChatGPT, Gemini) via its MCP.

Canva announced two cost-efficient versions of the model โ€” Craft-fast and Craft-ultrafast โ€” to be released soon (they were not yet released at launch).

Classification
Image generation
Family: Canva Craft
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
๐Ÿ“ฅ Input: text

Technical specification

License
Proprietary
Modalities
โฌ‡ Input
text
โฌ† Output
image

Capabilities and applications

Native model capabilities
Text-to-image generation
A model's ability to create images from a text description (prompt), including control of style, composition, aspect ratio and rendering text within the image.
Category: vision
Reasoning-augmented image generation
Multi-step generation: the model first analyses the prompt (planning layout, looking up web context, choosing references) and only then produces the image. This lets it accurately reflect complex, ambiguous instructions.
Category: reasoning
Text rendering in images
Generating images containing legible, correctly spelled text โ€” infographics, posters, menu cards, QR codes, captions in a specific graphic style. A key capability that separates new-generation models from early image generators.
Category: vision
Image editing
Modifying an existing image based on a text instruction or direct annotations: removing objects, changing style, adding elements, filling in regions (inpainting/outpainting), while preserving the identity of people and scene coherence.
Category: vision

Technical architecture

Model Form
TEText-to-Image

Sources and related pages