On Oct 29 Canva launched Brush, Canva Research's first vision-language model for design. It reads a design's layout, layers and style, edits anything from one line of copy to a whole deck, and iterates through an agent loop that writes code and then inspects the page. Canva says it is up to 69% faster and 46% cheaper per task than its previous design agent (vendor figures). It now powers supported design edits in Canva AI for Canva Pro subscribers.
Key Takeaways
- ✓Canva says Brush is up to 69% faster on general design editing tasks and 46% cheaper per edit task than its previous design agent (vendor figures from internal testing; sample size and conditions not published)
- ✓Brush was post-trained from a leading open-weights model and then refined with large-scale reinforcement learning, with rewards for visual quality, layout that follows design principles, and task completion
- ✓Built by Canva Research, a lab of about 160 people, one week after Canva shipped its first image model, Canva Craft
- ✓From Oct 29 it powers supported design edits in Canva AI for Canva Pro subscribers, selected from a Brush dropdown; wider availability is coming soon
- ✓The official post had about 276K views, 618 likes and 452 bookmarks when this was written
Key Decision Metrics at a Glance
Turn your technical choice into a development budget
Compare 40 dev plans & simulate token costs vs $20/mo subscriptions
Project Links & Resources
Direct AccessIn-Depth Technical Analysis
Background and pain point
Generating a poster with AI is easy; editing it is hard. A user asks to change one headline and the model also rewrites the colors, fonts and layout. Multi-page decks are worse, because titles, imagery and spacing must stay consistent from the first slide to the last. General-purpose models can reason through these edits but are expensive to run, and with hundreds of millions of people making billions of designs on Canva, a few extra cents per task adds up. On Oct 29 Canva announced Brush on its official account and newsroom as its first vision-language model for design.
How it works
Brush reads a design's layout, layers and style. Behind the scenes it runs an agent loop that writes and executes code to adjust design elements, then inspects the page to decide its next step. Text, images, colors and layout stay fully editable. It was post-trained from a leading open-weights model (Canva does not say which) on real design and editing tasks plus anonymized data from users who opted in, then refined with large-scale reinforcement learning whose definition of good design is set by professional designers on the team.
Benchmarks and results
Vendor figures: compared with Canva's previous design agent, Brush is up to 69% faster on general editing tasks and 46% cheaper per task, and in internal testing it matched or beat the old setup on most edits, especially restyling and resizing. Canva has not published public benchmark scores, sample sizes or the models it compared against.
For developers and users
Canva Pro users open an existing design, or attach or link one in Canva AI, and pick Brush from the dropdown. It suits three jobs: restyling an image or design, changing headers across a multi-page design in one prompt, and resizing a slide into another format such as an Instagram graphic. Brand Kit support, referencing attachments and better handling of image-heavy layouts are still in development.
Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.