# Canva launched its own AI model for creating images

*Canva introduced its own AI model, Craft, and announced high results in the Design Arena ranking. The model ranked fourth among image generation solutions, and in the categories of portraits and landscapes, it took first place.*  

[Canva Craft](<https://www.canva.com/dream-lab>) is the first proprietary artificial intelligence model for image generation developed by Canva Research. The model creates images with a resolution of up to 2K, supports more than 30 aspect ratios, and allows for more precise control over composition, object placement, and text.

Craft uses a two-stage architecture. In the first stage, the model analyzes the request and creates an image plan, determining its elements, their placement, sizes, visual hierarchy, and text. In the second stage, a separate model transforms this plan into a finished image.

This approach allows for the creation of complex compositions with readable text and accurate object placement. Additionally, the Craft architecture makes it possible to modify individual elements without affecting the rest of the image.

**Training on templates and quality assessment**

Unlike many image generation models that are primarily trained on photographs, Craft is oriented toward design. For training, millions of templates from the Canva library were used, including posters, social media posts, presentations, infographics, menus, and other materials with large amounts of text.

After pre-training, the model was further optimized using supervised fine-tuning and reinforcement learning. Professional Canva designers participated in the process, evaluating the quality of the images, the accuracy of instruction execution, and text rendering.

According to Canva, at the time of its launch, Craft ranked fourth in the Design Arena ranking among image generation models. However, in individual categories, it showed higher results: first place in the generation of images of people and portraits, as well as landscapes and nature. In the category of abstract images and patterns, the model took second place, and in the category of product images, fourth.

**What capabilities does Craft offer**

The model is designed for various creative tasks. With its help, you can create marketing materials, advertising banners, social media posts, realistic images of people and objects, illustrations, digital art, landscapes, and architectural visualizations.

Craft supports requests in different languages and can generate text in 55 languages. Users can choose from more than 30 aspect ratios or set a custom format depending on the task.

Integration with the Magic Layers feature allows converting generated images into editable design elements. This makes it possible to change individual parts of the image and adapt the composition to specific tasks.

Canva also worked on generation speed. The company used model distillation and developed its own request processing mechanism optimized for the image planning stage. In the future, Canva plans to release Craft-fast and Craft-ultrafast versions, which should provide faster generation at a lower cost.

**Where is the model available**

Canva Craft is already available to Canva users worldwide through the Dream Lab tool. Developers can also use the model via a public API for integration into their own products, including marketing platforms, e-commerce services, and advertising creation tools.

In the future, Craft is planned to be integrated with the Canva MCP Server, which will allow connecting the model to AI assistants and automated workflows.

Like other Canva AI products, Craft operates using the Canva Shield security system. It includes moderation of input requests and generation results, privacy and security control tools, and the ability to report potentially harmful or misleading content.

Canva Craft is the first version of the model, and the company plans to continue working on improving image quality, detail, and the accuracy of generating small text and technical drawings.
