Canva has unveiled Craft, its first proprietary artificial intelligence model for generating images. The main innovation is its ability to write text correctly in 55 languages, addressing one of the most common issues in AI visual generation tools.
The launch comes at a time when major tech companies are competing to offer increasingly precise image generators. Until now, Canva relied on third-party models for this function. With Craft, the company seeks to have full control over its own technology.
A model trained on designs, not photographs
Canva Research, the company's internal research team, developed Craft from scratch. The key lies in the training data: instead of using large volumes of photographs, the model learned primarily from millions of the platform's own design templates.
This approach directly influences how it works. Before generating the final image, Craft first plans the composition, proportions, and placement of the text. Only then does it build the visual result, which can reach 2K resolution and adapt to more than 30 formats, from square social media posts to landscape posters.
Magic Layers: edit by layers without starting from scratch
One of the most practical features accompanying Craft is Magic Layers. This tool separates the generated result into independent, editable layers.
This means that if the title of an image is good but the background is not, it is possible to modify only the background without touching the text, or vice versa. Until now, many users of AI image generators had to resort to external editors or regenerate the entire image to make small changes.
Availability and future plans
Craft is now available to all Canva users through Dream Lab, the section of the platform dedicated to AI generation. It can also be used via the public API, intended for developers to integrate the tool into their own applications.
Canva has also announced that it plans to integrate Craft later with conversational assistants like ChatGPT and Gemini. In the medium term, the company expects to launch two additional versions, called Craft-fast and Craft-ultrafast, aimed at offering faster and more economical results.
Why Canva is betting on a proprietary model
Having its own technology allows Canva to have greater control over the quality and speed of generation, in addition to reducing costs compared to using external models, according to the specialized media outlet The Decoder.
Researchers from Leonardo.ai, the AI image generation company that Canva acquired in 2024 and which was already part of its artificial intelligence integration strategy, participated in the development of Craft.
Comparisons with other giants, with nuances
Canva claims that Craft competes directly with image generation models from Google, OpenAI, ByteDance, and Meta. To support this claim, the company cites its position in the ranking of the DesignArena platform, where Craft placed fourth in the overall classification and first in the portrait and landscape categories.
It should be noted that these results come from Canva itself and not from an independent analysis, so its real performance against the competition remains to be verified externally.
What changes for those who use Canva
For those who use Canva on a daily basis, whether for business posters, greeting cards, or personal projects, Craft offers an integrated image AI that promises to generate legible text in the chosen language and deliver results that can be adjusted by layers.
This can avoid the step of switching between different applications or manually correcting AI-generated text, one of the most common weaknesses of these tools until now.
For professionals working with clients from different countries, the compatibility with 55 languages is especially relevant. It remains to be seen if the text accuracy is maintained in daily use just as it is shown in the company's demonstrations.
Frequently Asked Questions
What is Craft, the new Canva AI?
Craft is the first proprietary AI image generation model developed by Canva. It allows for the creation of images with integrated text in 55 languages.
How does Craft differ from other AI image generators?
It was trained primarily on design templates instead of photographs, and it plans the composition and text before generating the final image, which aims to improve text legibility.
What is Magic Layers?
It is a feature that separates images generated by Craft into editable layers, allowing for the modification of elements such as the background or text independently.
Where can I use Craft?
It is available to all Canva users through Dream Lab and also via the public API for developers.
Is Craft really better than Google, OpenAI, or Meta models?
Canva claims that Craft competes with those models and cites a ranking from the DesignArena platform where it placed fourth overall and first in portraits and landscapes, although these data are presented by the company itself, not an independent analysis.
Will Canva launch more versions of Craft?
Yes, the company plans to launch Craft-fast and Craft-ultrafast soon, which are faster and more economical versions of the model.
