Canva Ships Its Second AI Model in a Week: Brush Reads a Design's Layers, Edits 69 Percent Faster and Costs 46 Percent Less

Seven days after introducing Canva Craft, its first image-generation model, Canva is back with the other half of the plan. The company on Friday announced Canva Brush, its first vision-language model for design editing - a model trained to look at a design the way a designer does, understand how it is assembled, and change it on request without breaking everything else on the page.

The scope of what Brush handles is deliberately broad: a single line of copy, an image restyled to match a new palette, one change applied across every slide in a deck, or a design remixed from a presentation slide into an Instagram graphic. Canva's own example is a three-page bakery pitch deck in an Art Nouveau style, where the model plans the narrative across pages, generates and positions text, selects imagery, and keeps typography, colors and spacing consistent throughout.

Under the hood, Brush runs an agentic loop rather than a single-shot generation. It writes and executes code to adjust design elements, then inspects the actual result on the page to decide its next move, iterating until the task is done - an approach Canva compares to continuous self-evaluation instead of a series of isolated decisions. Crucially for a platform built on templates, every element Brush touches remains fully editable by hand, just like any other Canva design.

The model did not start from scratch. Canva says it fine-tuned a leading open-weights model, which it has not named, on real design and editing tasks, plus anonymized insights from users who opted in. The heavier lift was reinforcement learning: Canva built a reward system that scores each attempt on visual quality, adherence to design principles, and whether the requested task was actually completed, with the company's own professional designers setting the bar.

The performance numbers are Canva's own internal testing, so treat them as company claims until independent reviews arrive: Brush is up to 69 percent faster on general design editing tasks than Canva's previous design agent, and 46 percent cheaper per edit task. Canva matched or beat the old setup's quality on most editing tasks in internal tests, with the strongest gains in restyling and reformatting. The economics matter at Canva's scale - more than 220 million monthly users generating billions of designs - especially since the company slowed parts of its AI rollout earlier this year because the per-task costs were not sustainable.

Availability starts today: Brush powers supported design edits in Canva AI for Canva Pro subscribers, who select it from the model dropdown; wider rollout and Brand Kit support are promised next. The competitive framing is hard to miss. Adobe has acknowledged that layer-based AI editing is on its Firefly roadmap, but Adobe's current tools operate conversationally on single images, while Brush works across whole multi-page designs. Canva's bet is that specialized design models - Craft to generate, Brush to refine - will beat general-purpose models on both quality and cost, and the company says it will keep combining partner models with its own.