What is new
Canva has released its first in-house AI image model, called Canva Craft. Until now, Canva's AI features largely relied on partner models; Craft is built by Canva itself and is positioned as a direct competitor to image generators from OpenAI, Microsoft and Google. It is available now to all Canva users globally, accessible through the Dream Lab option in the platform's left-hand menu.
The headline feature is a two-stage architecture: rather than generating an image in one pass from a text prompt, Craft first plans the design and then renders it. Canva says this produces 'greater precision, consistency and control', particularly for images that need legible text and objects placed in specific spots, something that has historically been a weak point for AI image generators.
How it works
Most text-to-image models work in a single step: they take a prompt and a lot of learned visual patterns, and generate pixels directly, often struggling with coherent text, precise object counts or exact layout because there is no intermediate 'blueprint' guiding the composition. A two-stage approach, as described for Craft, separates planning from rendering. The idea is that by first deciding on a structure or layout for the design, the model has a clearer target to generate towards, which should reduce errors like garbled text or misplaced elements in complex scenes.
Canva says Craft was trained on millions of human-made templates drawn from its own content library, covering formats such as social media posts, presentations, infographics and menus. That training base is notable because it ties the model closely to the kinds of practical, structured design work Canva's users already do, rather than general-purpose photorealistic imagery, though Canva also says the model can produce realistic scenes, digital art and environmental imagery.
- Generates images at up to 2K resolution
- Supports more than 30 aspect ratios
- Trained on millions of human-made templates from Canva's library (social posts, presentations, infographics, menus)
- Currently ranked fourth on Design Arena's crowdsourced leaderboard for AI image generation models, behind OpenAI's GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT-Image-2
- Includes input and output moderation, privacy and safety controls, and reporting tools
- Available now globally via Canva's Dream Lab feature
Why it matters
For practitioners who use Canva for marketing, social content or quick visual assets, an in-house model means the design planning and image generation happen inside the same workflow Canva already offers, rather than routing through a third-party model with different quirks and limitations. The emphasis on precise text rendering and object placement addresses a common complaint about AI-generated graphics: that they look good at a glance but fall apart under scrutiny, with warped letters or oddly positioned elements. If Craft genuinely improves on that, it narrows the gap between AI-generated drafts and finished, usable designs.
Craft also signals where Canva wants to sit in the wider AI ecosystem. The company plans to let AI agents connect to Craft through the Model Context Protocol, an open standard for connecting AI systems to external tools and data. That would let automated agents (for example, a marketing workflow tool) generate or edit Canva designs programmatically rather than only through the Canva interface. Canva also says it plans faster variants, Craft-fast and Craft-ultrafast, suggesting the company intends to offer trade-offs between generation speed and output quality down the line.
Limits and open questions
The independent benchmark cited, Design Arena, currently places Craft fourth, behind three OpenAI models (GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare and GPT-Image-2). That is a reasonable starting position for a first in-house model, but it means Canva is not yet claiming to lead on this particular leaderboard. The announcement does not say how Design Arena's crowdsourced ranking is calculated, how large the voting pool is, or how rankings might shift as models update.
It is also not clear from the announcement how Craft's two-stage planning process is implemented technically, what the underlying model architecture is, how long generation takes compared with rival tools, or what the training data licensing looks like beyond the description of 'millions of human-made templates' from Canva's own library. Canva mentions moderation and safety controls but does not detail what these cover or how they are enforced in practice.
How to try it
Craft is live now for all Canva users. To access it, open the three-dot menu on the left-hand side of the Canva editor and select Dream Lab, where the image generation tools, including Craft, are housed alongside Canva's other AI design features such as prompt-based editing and layered AI-generated images introduced earlier this year.
Comments (0)
No comments yet. Be the first to share your thoughts.