Canva has built its own image model, and it is already in the hands of every user. The design platform announced on October 22 that Canva Craft, the first image generation model developed from scratch by its in-house Canva Research lab, is live for all users inside the Dream Lab tab and through the company’s public API. Canva says the launch makes it the world’s second-best AI lab for image generation on DesignArena’s public leaderboard, behind only OpenAI.
Until now, Canva has leaned on other companies’ models for image generation: Dream Lab launched in October 2024 running on Phoenix, the model from Leonardo.ai, the Sydney start-up Canva acquired that same year. Researchers from Leonardo worked on Craft, Canva told TechCrunch, which broke the launch news on October 22.
How Craft Works: Plan First, Then Paint
Most image models turn a prompt straight into pixels. Craft splits the job into two stages. First, a planning model reads the prompt and writes out a detailed blueprint of the composition: which elements appear, where they sit in relation to each other, their relative size, what matters most visually, and any text that has to appear in the image. Then a separate rendering model turns that plan into a finished picture at up to 2K resolution, in more than 30 aspect ratios or a custom size.
Canva says the planning step is what makes Craft different. Writing the layout before drawing gives legible text — the model can write text into images in 55 languages — objects placed where you asked, and edits to one element without redrawing everything else. The layered structure feeds directly into Magic Layers, Canva’s tool that turns a flat generated image into separate, editable design elements.
Trained On Designs, Not Photographs
Craft’s training data is what sets it apart from the photo-heavy datasets behind most rival models. Canva trained it on human-designed templates from its own library: posters, social posts, presentations, infographics, menus and typography-rich layouts. That design-first training means generated images come out structured like real design work, and can be converted into layered designs through Canva Magic, the company’s image-to-design transformer.
Last year Canva released a model that understands designs and layers; Craft completes the loop by generating them. The company also plans to support image generation inside third-party chatbots such as ChatGPT and Gemini through Canva’s MCP connector, turning its design engine into a tool other assistants can call.
Two More Versions Are Coming
Canva said it is already developing two cost-efficient versions of the model, Craftfast and Craft-ultrafast, which will be released soon — a signal that the company sees generation volume as the real game, not just flagship quality.
The business logic is straightforward. Canva has been paying other companies for generation on its platform, and owning the model stack trims that third-party API spend while giving it quality and speed control. The company cut its 2026 revenue growth target from 30 percent to 20 percent earlier this year, and its valuation reportedly slipped from $42 billion to $35 billion according to external analysts cited by TechCrunch — which makes cost control and a licensing channel for outside developers doubly attractive.
Why It Matters
Craft is the first serious attempt to build an image model that thinks like a designer instead of a photographer. Every major image model today — from Google’s, OpenAI’s, ByteDance’s and Meta’s — is trained to make beautiful pixels. Craft is trained to make editable work: layouts with a plan, text you can read, elements you can move. For the hundreds of millions of people who design in Canva, that is the difference between an illustration and a starting point they can actually finish.
The competitive signal matters too. Canva is pitching directly against the image models from Google, OpenAI, ByteDance and Meta, and its DesignArena ranking claim — second-best lab for images, behind only OpenAI — puts the whole industry on notice that a design company now believes it can out-build the model labs at their own game, at least in the category where design intent matters most.
Background
Canva, founded in Sydney in 2013, has grown into a design platform used by more than a quarter of a billion people, and it has been assembling AI pieces for years: the Leonardo.ai acquisition in 2024 brought media-generation researchers in-house, Dream Lab gave users a generation surface, and a design-understanding model shipped last year. Craft is the first public release from Canva Research as a standalone model builder — and it lands in a crowded stretch of model news, joining this week’s launches like iFlytek’s Spark X2.6, MiniMax’s M3.1 Flash and IBM’s open Granite Speech 5 in a week where every lab is shipping.
Source: TechCrunch ; Unite.AI
0 Comments