Canva Ships First Image Model, Craft, to Cut API Costs

Canva has released its first in-house image generation model, Canva Craft, to compete with models from Google, OpenAI, ByteDance, and Meta. The company built it with its own research team and made it available to all users through the Dream Lab tab on its platform.

The move matters commercially. Canva has relied on a range of third-party image models for generation on its platform, so an in-house model should cut its API costs. The company said the new model also gives it better control over quality and speed.

How does Craft generate images?

Canva said the model works in two stages. In the first stage, it interprets the user's prompt and builds a plan that includes the spatial relationships of elements, relative scale, visual hierarchy, and information for text generation. In the second stage, the model renders an image of up to 2K resolution in up to 30 different aspect ratios for various platforms.

The company said this two-stage approach produces images with accurate details of objects and legible text. Users can also make precise edits without changing other parts of the image.

That editability is rare in the category. Most consumer image generators rebuild the whole picture when a user asks for one change, and text inside images often comes out garbled.

Why train on templates instead of photos?

Canva made one unusual choice: it trained Craft on design, not photographs. The training data came from human-designed templates in Canva's library, including posters, social posts, presentations, infographics, menus, and typography-rich layouts.

That decision connects the model to the rest of Canva's product. Images generated with Craft can be converted into layered designs using Canva Magic, the company's image-to-design transformer released last year that understands designs and layers. A generated poster arrives as editable layers, not a flat picture.

Where can developers use it?

Beyond the consumer platform, Canva will offer Craft through the Canva API. The company also plans to support image generation inside third-party chatbots such as ChatGPT and Gemini through Canva's MCP.

The Leonardo.ai acquisition in 2024 supplied part of the talent. Canva bought the media generation platform last year and said Leonardo researchers contributed to the new model.

Two cheaper variants are next. Canva said it is developing cost-efficient versions called Craftfast and Craft-ultrafast, which will be released soon — a sign the company wants fast, cheap generation at scale across its product surface.

What does this mean for the business?

The launch lands at a delicate moment for Canva's finances. The company was on track for $4 billion in revenue last year. But it cut its growth forecast for 2026, and external analysts estimate its valuation fell from $42 billion to $35 billion.

Owning the image model instead of renting it fits that pressure. Every image generated by Craft instead of a third-party model removes an API fee, and the planned Craftfast variants push unit costs down further while giving Canva speed control it could not get from outside vendors.