Loading page…
Canva Craft plans a layout before generating an image. The company’s pitch centers on readable text, targeted edits and a path from generated pictures to layered designs.
Listen to this story
Canva is moving image generation deeper into its design stack with Canva Craft, its first in-house image model, now available to all users in Dream Lab as of October 22. Unlike a photo-first workflow, Craft plans layout, element placement and text before rendering. Canva says it supports 2K output and 30 aspect ratios, but its claims about readable text and precise edits have not been independently verified. The company says owning the model gives it more control over quality, speed and cost.
Canva says Craft was trained on human-designed templates—including posters, menus and presentations—rather than photographs.
Users can convert Craft images into layered designs with Canva Magic, a separate step from generating or editing an image.
Canva plans API access and image generation in third-party chatbots such as ChatGPT and Gemini through its MCP connection.
Canva users can now generate images with the company’s own model inside Dream Lab. Released October 22, Canva Craft is available to all users and brings a design-focused approach: plan the layout first, then render the image. Canva says that approach supports readable text and precise edits without disturbing the rest of a picture.
Craft works in two stages, according to Canva. First, it interprets the prompt and builds a plan. That plan covers where elements belong, their relative sizes, which parts should draw attention and information needed to generate text. Only then does the model produce the image.
Canva says the resulting images can reach 2K resolution, with support for 30 aspect ratios—the proportions between an image’s width and height. It also claims accurate object details, legible text and the ability to change a specific part without altering surrounding content. Those quality and editing promises are company claims, not independently verified results.
The training material follows the same design focus. Canva says it trained Craft on designs rather than photographs, using human-designed templates from its library. The examples include posters, social posts, presentations, infographics, menus and layouts rich in typography. Its stated training approach puts composed pages and their text at the center of the model.
Craft was developed by Canva’s research team, with contributions from Leonardo.ai researchers, the company said. Canva acquired the media-generation platform in 2024. The new release therefore draws on both its internal research operation and researchers from that acquisition.
Dream Lab is the current access point for Canva users. The company also says Craft will be available through its API, the interface developers use to connect software to a service. Beyond Canva’s platform, it plans image-generation support in third-party chatbots, including ChatGPT and Gemini, through its MCP connection.
Canva says an in-house model gives it more control over quality and speed, alongside costs.
Loading discussion...
Join the conversation
Explain when editing individual elements would matter to you.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.