Canva unveils Brush, a vision-language model for design

In its announcement, Canva writes that generating high-quality designs is one of the biggest challenges for AI models, but editing them is even more complex: a model has to understand what the user wants changed, how the design is structured, and what needs to stay exactly as it is. Otherwise, Canva says, a simple edit request can result in changes the user did not ask for.

Canva says Brush sees a design, understands how it is put together, and makes the requested changes. In its launch post on X, Canva says Brush reads a design's ‘layout, layers and vibe’ and can also build from a blank page.

According to Canva, Brush uses an agentic loop: it writes and runs code to adjust design elements, then inspects what is on the page to guide its next step, so it can keep evaluating and improving its work.

Canva illustrates the chain of decisions involved with the prompt “make me a 3-page pitch deck for my bakery with art nouveau visuals and backgrounds.” The company says an agent has to plan a narrative across pages, generate and place text, select imagery, and keep typography, colour, and spacing consistent from the first page to the last. Canva says everything Brush creates stays fully editable, like any other Canva design.

Canva lists three uses at launch: restyling an existing design or image, for example by adjusting the colour scheme, changing the weather, or describing a different vibe; making one sweeping change across a multi-page design instead of editing pages one by one; and repurposing a design for a new format, such as resizing a presentation slide into an Instagram graphic.

Canva says Brush started from a leading open-weights model, which it does not name, and was post-trained on real design and editing tasks using design expertise and anonymised insights from people who opted in to help improve Canva AI, followed by large-scale reinforcement learning. Its reward system, Canva says, scores visual quality, how well the layout follows design principles, and whether the task was completed as requested, and draws on Canva's design library, editing data, and hundreds of thousands of simulated interactions.

Canva reports that Brush is up to 69% faster for general design editing tasks and 46% lower in cost per edit task than its previous design agent, and that in internal testing it matched or exceeded the previous setup's quality on most editing tasks, with its strongest results in restyling designs and adapting them to different formats. These are company figures from internal testing, not independent benchmarks.

Canva says that starting October 29, Brush powers supported design edits in Canva AI for people on Canva Pro plans. Users open an existing design in the editor, or attach or link a design in Canva AI, and select ‘Brush’ from the dropdown list in Canva AI. Canva says it is working on Brand Kit support, the ability to reference attachments, and better handling of image-heavy layouts.

Brush follows Canva Craft, Canva's first image-generation model, released October 22. Canva says Craft generates the image and Brush refines the design around it, and that both come from Canva Research, its 160-person AI lab. Canva says Craft debuted as the fourth-ranked image generation model on DesignArena and made Canva the second-ranked AI lab for image generation on that leaderboard.