Canva Brush Arrives: The Vision-Language Model That Edits Your Existing Designs by Prompt

What Happened: Brush Went Live Today

On October 30, 2026, Canva announced Brush, a new in-house vision-language model built by its Canva Research team to power the next generation of the company's AI design features. Unlike most AI design releases this year, Brush is not about generating images from scratch — it's about editing and refining work that already exists.

According to Canva, Brush understands the structure of an existing design and makes requested changes while keeping everything else intact. The scope ranges from a single line of copy to an entire multi-slide presentation. In Canva's own example, you can ask it to create a three-page pitch deck for a bakery with Art Nouveau visuals — and the model plans the narrative across pages, generates and positions text, selects imagery, and maintains a consistent visual style throughout.

Brush is Canva's second major in-house model push, following Canva Craft, the image-generation model it built to reduce reliance on third-party providers. Together they signal the same strategic move we've watched OpenAI, Google, and Adobe make all year: when you operate at Canva's scale, renting someone else's model for your core product becomes both a cost problem and a differentiation problem.

How It Works: An Agent That Writes and Runs Code

The most technically interesting detail is how Brush edits. Rather than regenerating pixels — the approach most image models take — Brush uses an agentic process that writes and runs code to adjust design elements, then inspects the resulting page to decide what to change next. It's a code-first edit loop applied to a visual canvas.

That design choice matters for three reasons:

• Consistency across pages. Brush was explicitly built for tasks that require typography, colors, spacing, imagery, and layout to stay coherent across an entire document — the thing pure image generators are worst at.
• Everything stays editable. Because edits are structural operations on design elements rather than pixel output, users can still modify text, imagery, colors, and layouts manually at any time. Your design remains a design, not a flattened image.
• Verifiability. The inspect-after-edit loop lets the model catch its own mistakes — a page that drifted off-style gets corrected in the next step instead of shipping broken.

It's the same "AI writes code to act on the world" pattern that reshaped coding tools in 2026, applied to design software instead. The canvas becomes an environment, and the model becomes an agent operating in it.

The Numbers: 69% Faster, 46% Cheaper per Edit

Canva's claims about Brush versus its previous design agent:

Those are vendor-reported numbers from Canva's own internal testing, so treat them as directional rather than gospel. But the cost-per-edit figure is the one to watch: for a platform serving hundreds of millions of users, halving the marginal cost of every AI edit is the difference between AI features being a margin drain and a profit center. It also explains why Canva built the model itself instead of continuing to buy API capacity from frontier providers.

Generation Is Solved; Editing Is the New Battleground

Brush lands on a market that has quietly shifted. For two years, AI design tools competed on generation — who could make the prettiest image from a prompt. In 2026, the competition moved to editing existing work, because that's where real design time actually goes:

• Restyling an old deck for a new brand, page by page, is tedious, repetitive, and perfectly suited to an agent that maintains consistency.
• Reformatting a design across sizes and formats — a poster into a story, a story into a banner — is exactly where Canva says Brush performs strongest.
• Iterative refinement (“make the headline shorter, keep everything else”) has always broken image generators, which regenerate the whole picture and lose your layout. Structural editing fixes that.

Adobe has been making the same bet from the professional end with Adobe Firefly and generative editing inside Creative Cloud, while vector-native tools like Recraft have pushed AI editing of editable artwork specifically. Canva's advantage is distribution: it's bringing structural AI editing directly to the tens of millions of non-designers who already live in its editor.

How Brush Fits Among the AI Design Tools

If you're picking a tool today, here's the rough map:

• Canva + Brush — best for non-designers and teams who need to edit and restyle existing templates, decks, and social graphics at speed. Requires Canva Pro.
• Adobe Firefly / Creative Cloud — best for professionals who need generative editing inside a full design suite, with commercial-safe training data as the selling point.
• Recraft — best for AI-generated editable vector art and brand-consistent image styles.
• Midjourney and Ideogram — still the strongest pure image-generation aesthetics; use them to create assets, then edit elsewhere.
• Figma — the product-design standard, where AI features increasingly compete for professional UI workflows.
• Beautiful.ai — presentation-first AI for teams that live in slides.

The honest summary: no single tool wins everything. A common 2026 stack is generation in one tool, structural editing in another, and delivery in your team's editor of choice.

How to Try It

Brush is rolling out now for Canva Pro subscribers for supported design edits inside Canva AI. To use it: open an existing design in the editor (or attach or link a design inside Canva AI), then select Brush from the model dropdown menu in Canva AI. Free-plan users will likely wait for a wider rollout, as with previous premium AI features.

If you're comparing plans and pricing across AI design tools — including Canva's — you can find current details and alternatives for each on aitrove.ai.

Frequently Asked Questions