Canva has unveiled its first internally developed artificial intelligence image generation model, named Canva Craft. Built by the company's own research team, the model positions the design platform as a direct competitor to similar offerings from major technology companies including Google, OpenAI, ByteDance, and Meta.

The announcement marks a significant step for Canva, which has historically relied on third-party AI models to power image generation features within its platform. With Canva Craft, the company is now entering the space with a proprietary solution developed from the ground up.

How the Model Works

According to Canva, Canva Craft operates through a two-stage process. In the first stage, the model analyzes a user's text prompt and constructs a structured plan that accounts for the spatial relationships between visual elements, their relative scale, visual hierarchy, and any requirements related to text generation within the image.

In the second stage, the model produces the actual image based on that plan. Outputs can reach up to 2K resolution and are available in as many as 30 different aspect ratios, making the tool adaptable to a wide range of publishing platforms and use cases.

Canva says this two-stage architecture is what allows the model to render objects with accurate detail and produce legible text within images - a challenge that has historically been a weak point for many AI image generators. The approach also enables users to make targeted edits to specific parts of an image without inadvertently altering other elements.

Trained on Designs, Not Photographs

One of the more distinctive aspects of Canva Craft is how it was trained. Rather than building its dataset from photographs - as is common with many image generation models - Canva trained Craft using human-designed templates sourced from its own extensive library. These include posters, social media posts, presentations, infographics, restaurant menus, and layouts that are heavily reliant on typography.

This design-centric training approach has a practical downstream benefit. Images produced by Canva Craft can be converted into fully editable, layered design files using Canva Magic, the company's existing image-to-design conversion tool. This workflow could prove particularly useful for designers who want to use AI-generated visuals as a starting point and then refine them within Canva's broader editing environment. The company had already introduced a dedicated model last year capable of understanding design structures and individual layers.

Availability and Integration

Canva Craft is now available to all users on the platform. It can be accessed through the Dream Lab tab within the Canva interface. In addition to direct platform access, the model will be made available through the Canva API, opening it up to developers and businesses building on top of Canva's infrastructure.

The company also has plans to extend Canva Craft's reach through third-party AI assistants. Specifically, Canva intends to support image generation in tools like ChatGPT and Google Gemini via Canva's Model Context Protocol, commonly referred to as MCP.

Cost Savings and Performance Control

Before developing its own model, Canva had been integrating a variety of external image generation models into its platform. By replacing or supplementing those with an in-house solution, the company is expected to reduce its reliance on third-party API costs. Canva acknowledged this directly, noting that beyond cost efficiency, the proprietary model also gives the company greater control over output quality and processing speed.

The company is also developing two lighter versions of the model aimed at faster performance and lower computational overhead. These variants, named Craft-fast and Craft-ultrafast, have been announced but have not yet been released. Canva indicated they are expected to launch in the near future.

Leonardo.ai's Contribution

The development of Canva Craft also reflects the company's 2024 acquisition of Leonardo.ai, a media generation platform known for its AI-powered creative tools. Canva confirmed that researchers from Leonardo.ai contributed meaningfully to the development of the new image model, suggesting the acquisition has begun to yield tangible technical returns.

Financial Context

The launch comes at a complex moment for Canva financially. The company was reported to be on course to reach $4 billion in annual revenue in 2024, a milestone that reflected strong growth across its user base and enterprise offerings. However, the company subsequently revised its growth projections for 2026 downward, a move that drew attention from outside analysts.

Following that revision, Canva's estimated valuation was reported to have dropped considerably - from approximately $42 billion to around $35 billion according to external assessments. Despite that recalibration, the company continues to invest in expanding its core product capabilities, with the introduction of Canva Craft representing one of its most substantial technical undertakings to date.