For the past year, Canva has worked to become what it calls a creative operating system: a single platform where AI, design, and productivity come together. To make that idea a reality, the company has been developing its own models to handle everything from style transfer to video generation faster and more cheaply than frontier alternatives. But people come to Canva for more than layouts—they use it to create images, so the company launched Craft this week, which it described as its first image-generation model, built and trained in-house from scratch.
Canva’s Chief Executive Melanie Perkins once said that the company’s mission “has always been to empower the world to design” and that design wasn’t about having something look pretty. “[It’s] about giving form to an idea.” Craft is designed (no pun intended) to help creators do just that. After all, it’s one thing to identify what you want to create, but how it looks can be just as complicated.
“We’re incredibly proud of what we’ve achieved with Craft, and how it blends design understanding with best in class image generation,” JJ Fiasson, Canva’s head of generative AI, wrote in the company’s announcement. “Craft delivers exceptional results for jobs where the layouts and text need to be perfect, and renders people and landscapes with stunning quality.”

Trained on millions of human-made templates—posters, social posts, presentations, infographics, menus, and typography-rich layouts—from Canva’s extensive content library, Craft can generate photorealistic images in 2K resolution across more than 30 aspect ratios. The company explained that by training the model on its templates, the AI learns how layouts, hierarchy, and typography fit together, giving it a better sense of what effective design looks like. Supervised fine-tuning and reinforcement learning further optimized Craft to improve the model’s ability to follow instructions, typography accuracy, and overall visual quality.
Craft isn’t the only image-generation model Canva has developed. It also has Lucid Origin, which was announced in April. However, that’s a byproduct of Leonardo.ai, a company Canva acquired in 2024. Craft is the inaugural model of its type to emerge from Canva Research.

While the company hasn’t revealed the new model’s size, technical paper, or API pricing, we do know Craft supports text creation across 55 languages. It can generate a wide range of images, from social posts and marketing assets to lifestyle and editorial photos, illustrations and digital art, people and characters, nature and environment imagery, and creative and conceptual work. Craft also appears to be a closed model and will eventually have two variations: Craft-fast and Craft-ultrafast designed to “deliver high-quality image generation at a fraction of the cost of leading frontier models.”
On performance, Canva cited findings from DesignArena. Craft reportedly took the top spot globally for People and Portraits, ahead of 86 other models, and for Landscapes and Nature, ahead of 84 other models. It placed second in Abstract and Patterns and fourth in Products. Canva said that when Craft debuted on DesignArena, it was the “fourth most advanced image generation model,” just behind OpenAI’s GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT Image 2. By this logic, Canva called its research team the “world’s second most advanced lab for image generation.”

Canva’s Craft model is available today for all users worldwide through its website (open Dream Lab in the bottom-left panel) or the public API. An MCP server is expected to launch soon that will let Craft connect with AI assistants and automated workflows.
Releasing an image model complements the company’s existing AI portfolio, but Canva will likely follow up with other generation models to complete its creative operating system vision. Regardless, Craft’s debut puts more pressure on Adobe to compete more aggressively in the creative and productivity space and comes weeks before Adobe’s annual Max conference and amid a leadership shakeup.
It’ll be interesting to hear what creatives and marketers think of Craft and how it compares to Adobe’s Firefly. Adobe has been investing heavily in its offering, releasing first-party image, audio, and video model generations, adding third-party model support, and integrating its creative app suite directly into AI assistants and workplace apps like Claude, Gemini, ChatGPT, and Slack. But that doesn’t mean everything is rainbows and sunshine—the platform has faced complaints about inconsistent results, pricing issues, and more.
Still, Adobe has built its AI pitch around ethics, saying it doesn’t train its Firefly models on customers’ work and positioning them as safe for commercial use. It said that it mainly used content from Adobe Stock, its database of hundreds of millions of licensed images. How Canva handles this is less clear. It hasn’t said whether it compensates the creators of the templates it used to train Craft or whether it will protect customers who use the images commercially. Instead, in the company’s announcement, it said that its models “undergo continuous testing and rigorous safety evaluations to identify and mitigate potential risks. Alongside visual quality, we evaluate brand safety, cultural context, and the potential for harmful content to ensure our models meet our standards before they’re released.”
Discover more from Ken Yeung
Subscribe to get the latest posts sent to your email.
