Canva Launches Craft, Its First In-House Image Generation Mode

Design platform Canva has released Canva Craft, its first in-house image generation model, a foundation model built and trained from scratch by the company’s research team. The launch marks a decisive shift for a platform that, until now, leaned largely on third-party models to power its generative features, and it drops Canva straight into a category already crowded with offerings from Google, OpenAI, ByteDance and Meta.

Unlike many rivals that optimise primarily for photorealism, Craft is designed around the things people actually make in Canva: posters, presentations, social posts, infographics and typography-heavy layouts. The company says the approach lets the model render accurate object details and legible in-image text, and lets users make precise edits without disturbing the rest of a design.

“With Craft, Canva says it now ranks as the world’s second most advanced AI lab for image generation, behind only OpenAI.”

A model that plans before it paints

Canva Craft splits image creation into two distinct stages. In the first stage, the model interprets the user’s prompt and writes out a plan of the composition, the spatial relationships between elements, their relative scale, the visual hierarchy of the layout, and the instructions needed to generate any text inside the image. In the second stage, it translates that plan into a finished image at up to 2K resolution, with support for roughly 30 aspect ratios tuned for different platforms and formats.

That separation of planning from rendering is what Canva credits for Craft’s stronger text handling and its more controllable, editable output. Because the composition is planned explicitly, generated images can also be converted into layered, fully editable designs through Canva Magic, the company’s image-to-design transformer, the same lineage as the design-aware model Canva introduced in 2025.

Canva Craft at a glance

The headline specifications position Craft as a general-purpose design model rather than a pure photo generator. Key details reported at launch are summarised below.

Table 1 · Key specifications

How Craft stacks up against the field?

Canva points to preliminary results on Design Arena, an independent, human-judged leaderboard for AI-generated design — where Craft placed fourth on the image-generation ranking. According to Canva, that result lands the model ahead of entries associated with Google, Microsoft, Meta, xAI and ByteDance, and makes Canva the second-highest-ranked image-generation lab behind OpenAI. Design Arena evaluates models through head-to-head human comparisons, using a Bradley-Terry rating system drawn from millions of user votes.

Table 2 · Competitive positioning (as reported at launch)

Note: Design Arena rankings are preliminary and community-driven, and shift as new models and votes are added. Figures reflect what Canva and reporting outlets stated at launch.

Years in the making

Craft does not arrive in a vacuum. It is the latest step in a multi-year build-out of Canva’s AI stack, anchored by its 2024 acquisition of media-generation platform Leonardo.ai, whose researchers contributed to the new model, and by the design-aware foundation model the company shipped in 2025 to power features such as Magic Layers and Canva AI 2.0.

Faster, cheaper versions on the way

Canva says Craft is only the first release in a family of models. Two cost-efficient variants — Craftfast and Craft-ultrafast are in development and expected soon, aimed at trimming latency and cost for high-volume use.

Table 3 · The Craft model family

Owning the model to control the bill

Bringing image generation in-house is as much a business move as a technical one. Canva has run a mix of third-party image models across its platform, so a model it owns gives it tighter control over quality, speed and, crucially, cost. The company has not disclosed exactly how much it expects to save, but the economics matter: Canva was on track for roughly $4 billion in revenue, yet trimmed its 2026 growth forecast as it absorbed high AI costs, a move external analysts linked to a reported dip in its valuation from about $42 billion toward the mid-$30 billions.

Exposing Craft through a public API and an MCP server also opens a new commercial lane: Canva can position its model against the image endpoints sold by OpenAI and Google, courting product teams building marketing platforms, e-commerce sites, advertising tools and print services.

Where to try it

1. In Canva: available to all users from the Dream Lab tab.
2. For developers: accessible through Canva’s public API.
3. In other chatbots: image generation inside ChatGPT and Gemini via Canva’s MCP.

The bottom line

With Craft, Canva stops being purely a consumer of other companies’ image models and becomes a producer in its own right. The design-first training and plan-then-render architecture give it a genuine point of difference in a field obsessed with photorealism, and the API and MCP hooks turn an internal cost-saver into a potential revenue stream. Whether Craft holds its leaderboard standing as rivals iterate is the open question, but the direction of travel is clear: Canva now wants to be judged as an AI lab, not just a design tool.

Emily Carter

Writer · TechJox

Emily Carter is a technology writer at TechJox, specializing in Artificial Intelligence, emerging AI tools, software platforms, and digital innovation. She covers the latest tech news, cybersecurity updates, product reviews, and automation trends. Her writing focuses on research-driven insights, practical information, and simple explanations that help readers understand evolving technologies and stay informed about the digital world.