Canva releases its own Canva Craft image generation model — TechCrunch

Design platform Canva has introduced its first in-house image generation model, Canva Craft. It is already available to all users in the Dream Lab tab and is also expected to become available through the Canva API, TechCrunch reports.

The model was developed by Canva's research team. The company positions it as a competitor to image creation tools from Google, OpenAI, ByteDance, and Meta.

Generation in two stages

According to Canva, Craft works in two stages. First, it analyzes the user's prompt and creates an image plan: it determines the spatial arrangement of objects, their relative scale, visual hierarchy, and data for generating text. At the second stage, the model creates images with resolutions of up to 2K in 30 aspect-ratio formats.

Canva said this approach should ensure more accurate reproduction of object details and legible text in images. Users will also be able to edit individual elements precisely without changing other parts of the work.

More current news is available on the UA.News Telegram channel Telegram.

Training on design templates

Canva said it used human-created design templates from the platform's library, rather than photographs, to train Craft. The dataset included posters, social media posts, presentations, infographics, menus, and typography-rich layouts.

Images generated in Canva Craft can be turned into multilayered designs using Canva Magic, an image-to-design conversion tool. Last year, the company introduced a model that recognizes design and layers.

Canva also plans to support image generation in third-party chatbots, including ChatGPT and Gemini, through its MCP. In 2024, the company acquired the media generation platform Leonardo.ai; its researchers contributed to the creation of the new model. Canva plans to release two more economical versions, Craftfast and Craft-ultrafast, in the near future.

The company said its own model will help it better control the quality and speed of the service, as well as reduce spending on other models' APIs.