Canva launches its first in-house image generation model, available to all users for creating precise, editable designs with legible text.
Key points
- Canva Craft uses a two-stage process to plan layouts before generating images.
- The model is trained on human-designed templates rather than raw photographs.
- It is available to all users via the Dream Lab tab and through the API.
Canva has released Canva Craft, its first proprietary image generation model. The tool is now available to all users on the platform and through its application programming interface.
What happened
Design platform Canva launched Canva Craft, a new image generation model developed by its internal research team. According to TechCrunch, the model competes with offerings from Google, OpenAI, ByteDance, and Meta. Canva states that the system operates in two distinct stages to improve output quality.
The first stage interprets user prompts to create a detailed plan. This plan includes spatial relationships, relative scale, visual hierarchy, and text generation data. The second stage generates the final image based on this structured plan.
Canva Craft produces images at up to 2K resolution. It supports more than 30 different aspect ratios for various digital platforms. The company says this approach allows for accurate object details and legible text within the generated visuals.
Users can make precise edits to specific parts of an image without altering other areas. Canva notes that the model was trained on human-designed templates from its library, such as posters and infographics, rather than on photographs.
Researchers from Leonardo.ai, which Canva acquired in 2024, contributed to the development of the model. The company plans to release faster, cost-efficient versions called Craftfast and Craft-ultrafast soon.
Why it matters
AI image generation often struggles with complex layouts and text. Traditional models may produce distorted letters or misplaced elements because they predict pixels without understanding structure. Canva’s two-stage approach addresses this by separating planning from rendering. This allows the system to understand spatial relationships and visual hierarchy before creating the final image. For designers, this means more reliable outputs that require less manual correction. It shifts the tool from a creative assistant to a precise design engine.
The fix
- Review generated images for subtle distortions in text or complex objects.
- Use the layered editing features to adjust specific elements without regenerating the whole image.
- Test different aspect ratios to ensure the design fits your target platform.
- Monitor the release of Craftfast and Craft-ultrafast for improved performance.
Why it matters
Generative AI tools are moving beyond simple image creation toward structured design workflows. By training on templates rather than photos, models can better understand the rules of graphic design. This includes alignment, spacing, and typography. Teams using these tools should evaluate how well the AI respects layout constraints. It is important to verify that generated content aligns with brand guidelines and accessibility standards before publication.
What to do and how to stay safe: Canva
- Check the Dream Lab tab in Canva to access the new model.
- Compare outputs from Canva Craft with previous versions to assess quality improvements.
- Use the API integration if you need to automate image generation for large projects.
- Watch for announcements regarding the faster Craftfast and Craft-ultrafast models.
Step-by-step guide: What is password policies and how they protect systems
General security guidance from the Crypt Weekly newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How does Canva Craft differ from other AI image models?
It uses a two-stage process that plans the design layout before generating the image, allowing for better precision and legible text.
Is Canva Craft available to free users?
Yes, the model is available to all users on the Canva platform, accessible via the Dream Lab tab.
What data was used to train Canva Craft?
The model was trained on millions of human-designed templates from Canva’s library, including posters and infographics, rather than photographs.
