Canva releases its first image generation model

Canva has released its first image generation model, called Canva Craft, developed by its own research team. The company is positioning the model against same-category offerings from Google, OpenAI, ByteDance and Meta, and is making it available to all users on its platform through the Dream Lab tab.
The model works in two stages. First, it reads the user's prompt and builds a plan that captures details such as the spatial relationships between elements, their relative scale, the visual hierarchy, and the information needed for text rendering. Second, it generates an image at up to 2K resolution, in any of 30 possible aspect ratios suited to different platforms. Canva says this planning step is what lets the model produce images with accurate object detail and legible text, and lets users make precise edits without disturbing other parts of the image. That combination of editable text and localized edits is a common weak spot for image generators, so it is the claim worth testing first.
Training data is the other notable choice. Rather than photographs, Canva trained Craft on designs, drawing on human-made templates from its own library: posters, social posts, presentations, infographics, menus and typography-heavy layouts. This suggests the model is tuned toward the kind of graphic-design output Canva's users actually produce, rather than photorealistic imagery. A likely trade-off is that photo-style prompts may fare less well than layout-driven ones, though the company has not made such a claim either way.
Because the output is design-shaped, it can be handed to Canva Magic, the company's image-to-design transformer, and converted into a layered design. Canva released that model last year; it understands designs and layers, which is what makes downstream editing in Canva's editor possible. In practice, the two-stage generation plus a layer-aware converter is what separates a design-tool model from a general-purpose one: the generated file stays workable rather than becoming a flat bitmap.
Availability starts on Canva's own platform for all users via the Dream Lab tab. The model will also be offered through the Canva API, and the company plans to support image generation inside third-party chatbots such as ChatGPT and Gemini through Canva's MCP. For developers, the MCP route is the one to watch closely: it would let an assistant inside another product call Canva's generation without a direct API integration, which is a meaningfully different distribution path from a traditional SDK.
Canva says the in-house model does more than cut dependence on outside vendors. It points to better control over quality and speed, alongside lower costs. Canva has relied on a range of third-party image models inside its platform, so owning the stack plausibly reduces per-image API spend, a reasonable inference given the company's own emphasis on cost. Two cheaper variants are in development, named Craftfast and Craft-ultrafast, and are slated for release soon; no dates or prices were given.
The research lineup has history behind it. Canva acquired the media generation platform Leonardo.ai in 2024, and says researchers from Leonardo contributed to Craft. The release also lands against a mixed financial backdrop: the company was on track for $4 billion in revenue last year, but it cut its growth forecast for 2026, and its valuation reportedly fell from $42 billion to $35 billion, according to external analysts. Owning generation rather than licensing it fits a company under margin pressure.
Why it matters: design and product teams that already build on Canva get a native generator whose output can flow into layered, editable designs, which reduces the round trip between generating an asset and refining it. The API and MCP plans mean the model could show up inside third-party assistants, so teams integrating Canva should treat this as a new generation endpoint to evaluate rather than a cosmetic feature. The open questions are practical: output quality on photo-style prompts, the cost gap between Craft and the coming fast variants, and how reliably generated text survives edits at 2K across 30 aspect ratios.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Canva is making its new image model available to all users, promising better, more customizable performance at lower costs.