Canva Launches Canva Craft, Its First Image Generation Model
The design platform opens its new model to all users, promising better quality and lower costs as it cuts its 2026 growth forecast
By AXY AI Desk ·
Key points
- Canva has released Canva Craft, its first image generation model, to all users through its Dream Lab tab.
- The model works in two stages, planning the layout first and then generating images up to 2K resolution in 30 aspect ratios.
- Canva trained the model on human-designed templates rather than mainly on photographs, and plans to offer it through its API and third-party chatbots.
- Canva said the model should lower its API costs, and it plans two cheaper versions called Craftfast and Craft-ultrafast.
- The launch comes after Canva cut its 2026 growth forecast, with its valuation reportedly falling from $42 billion to $35 billion according to external analysts.
Canva has released its first image generation model, called Canva Craft, and is making it available to every user of its design platform. The model was built by the company's own research team to compete with image models from Google, OpenAI, ByteDance and Meta. Canva says it delivers better, more customizable results at lower costs, which matters to the millions of people who use its tools to make posters, social posts and presentations.
Users can reach the model from the Dream Lab tab inside Canva. The company also plans to offer it through the Canva API, a application programming interface, which is a way for other software to plug into a service and use its features. Canva said it intends to support image generation inside third-party chatbots such as ChatGPT and Gemini through Canva's MCP, a connector standard that lets AI assistants work with outside software. If that happens, someone chatting with an assistant could ask for a graphic and get a Canva-made image without opening the design app.
How Canva Craft Works in Two Stages
Canva describes the model as working in two steps. In the first stage, it reads the user's prompt, the written instruction describing what they want, and turns it into a plan. That plan includes the spatial relationships between elements on the page, their relative size, the visual hierarchy that tells a reader what to look at first, and the information that will go into any text.
In the second stage, the model generates the actual image. Canva says it can produce images up to 2K resolution, a little under 2,000 pixels across, in as many as 30 different aspect ratios. That range matters because a square Instagram post, a widescreen slide and a tall poster need different shapes, and designers usually have to resize work by hand for each one.
The company said this split approach lets the model get the details of objects right and produce legible text, which has been a weak point for many image generators. Text that comes out garbled or misspelled is one of the most common complaints about tools of this kind. If the plan is set first, the model has a clearer blueprint for where words and objects should go.
Why Canva Trained on Designs Instead of Photographs
One of the more unusual choices Canva described is what it trained the model on. Rather than relying mainly on photographs, the company said it used designs created by humans. Those included templates from Canva's own library: posters, social posts, presentations, infographics, menus and layouts built around typography, meaning the arrangement and styling of letters.
That choice is central to what the company is selling. A photograph teaches a model what things look like. A well-made template teaches it how things are arranged on a page, how headlines sit above body copy and how a menu balances prices against dish names. For a company whose customers mostly need finished layouts rather than art for its own sake, that kind of training could make the output more useful on the first try.
Precise Edits and Layered Designs
Canva says users will be able to make precise edits without changing other parts of an image. In a typical image generator, asking for a change to one corner can alter the whole picture, and people often end up regenerating until they get something close to what they wanted. Localized edits are a practical benefit for anyone fixing one word on a sign or swapping a product without losing the rest of a layout.
The company also said images made with Canva Craft can be converted into layered designs using Canva Magic, which it described as its image-to-design transformer. A layered design is one where the parts, such as text, shapes and photos, are separate objects that can be moved, recoloured or replaced. A flat picture cannot be edited that way. Canva noted that it released its own model last year that understands designs and their layers, so this feature builds on earlier work.
Why Canva Is Building Its Own Image Model
Canva has been using a number of image models from outside providers to power generation on its platform. The company said that a model it controls should help reduce its API costs, since it would no longer pay for each image generated by a third party in the same way. It also said the new model gives it more control over quality and speed, two things that can change how a product feels to someone waiting for a result.
The company has history in this area. In 2024, Canva acquired Leonardo.ai, a media generation platform. Canva said researchers from Leonardo contributed to the new image model, which suggests the acquisition is now feeding directly into products that customers use.
This is also a sign of where the design software market is heading. Companies that once bought image generation as a service are increasingly building their own, because owning the model means owning the cost structure and the quality bar.
Cheaper Versions and the Revenue Pressure Behind the Launch
Canva said it is developing two lower-cost versions of the model, called Craftfast and Craft-ultrafast, which it will release soon. The company did not give prices or performance figures for those versions in the material available, so the practical effect on what users pay is still unclear. Still, the naming makes the strategy plain: a full-quality model for complex work, and faster, cheaper options for high volumes.
The timing also comes against a difficult financial backdrop. Canva was on track for $4 billion in revenue last year. However, it cut its growth forecast for 2026, and external analysts cited in the reporting said its valuation had reportedly fallen from $42 billion to $35 billion. A slower growth outlook can push a company to find savings and to use new products to win back attention, and a cheaper in-house model fits both goals.
That pressure gives the model commercial weight beyond the technology. If Canva can cut the cost of each generated image while keeping quality high, it can offer more generation features to free and paying users without sending the bill up in the same proportion.
What Could Come Next for Canva Craft
The next test is delivery. Canva has promised third-party chatbot support, API access and the two cheaper variants, and each of those depends on execution. If the chatbot integration works smoothly, Canva could reach people who never open its app. If the cheaper versions arrive on time and keep quality, the company will have a stronger argument that building its own model was worth the cost.
Users will also watch how the layered design feature performs on real work. A clean conversion from a generated image into editable parts would make Canva Craft useful for brand teams that need to update materials quickly. Poor results, such as misplaced text or rough edges between layers, would limit its value for professional designers, who tend to be the most demanding users.
For now, the release puts Canva in direct contact with the largest names in image generation. Whether that contact turns into market share depends on how the model compares in everyday use, and on whether the company can hold its promised costs and speed as more people start making images with it.
Questions and answers
What is Canva Craft?
Canva Craft is the first image generation model from design platform Canva. It was developed by Canva's research team and is available to all users through the Dream Lab tab.
How does Canva Craft generate images?
The model works in two stages. It first builds a plan covering spatial relationships, relative scale, visual hierarchy and text, then generates an image of up to 2K resolution in as many as 30 aspect ratios.
Will Canva Craft be available outside Canva?
Canva says the model will be offered through its API and that it plans to support image generation in third-party chatbots such as ChatGPT and Gemini through its MCP connector.
Why is Canva building its own image model?
Canva said its own model should help reduce API costs and give it better control over quality and speed, after using several image models from other providers.
Topics: Canva, Canva Craft, Image generation, Leonardo.ai, Canva Magic, Generative AI, Design software
Sources: TechCrunch AI
https://axynewsglobal.com/article/canva-launches-canva-craft-its-first-image-generation-model