Image: The New Stack
Why it mattersA design model that keeps every element editable removes the flat-image problem. Teams can send a half-finished file to the model and get the same file back.
A designer who runs a model on a draft usually gets a flat image back and has to start again. Canva has released a design model that keeps the file editable. It writes and runs small pieces of code to adjust each element in place. Then it looks at the result and picks what to do next. Canva calls it Brush, and it is in Canva AI for Pro subscribers from 29 October.
Brush is a vision-language model built by the 160-person Canva Research team. Canva says it reads a design's content and layout. It writes code to change one element at a time. It then inspects the page before the next step. Every element it touches stays editable in Canva.
What Canva says it does
The model works in a loop. The New Stack cites Canva's own words. Brush writes and runs code to adjust the design. It then inspects the result before the next step. Canva also says the model keeps typography, color and spacing in step across a multi-page design.
Canva reports two numbers against its earlier design agent. Brush is up to 69 percent faster at general design-editing tasks. It is also 46 percent cheaper per edit. Both figures come from Canva's own measurements against a predecessor only Canva has run.
Why editing code is a different design
Most AI design tools generate a new image from a prompt. The output is pixels. A designer who wants to move a shape six points has to start over. Brush edits the file itself. A heading stays text, and a shape stays a vector. A client change on Monday can land in the same file on Friday.
Canva pairs Brush with another model called Craft, announced on 22 October. Canva says Craft generates the image, and Brush refines the design around it. Both come from Canva Research. The company says Canva AI brings together partner models and these in-house models for creative tasks.
The limits on the vendor numbers
Canva's 69 percent and 46 percent figures are self-reported against a predecessor. They cover the gap between Canva's two in-house models. No third party has run its own comparison with image-generation models from other vendors.
Access is also limited. Brush is in beta, inside Canva AI, for Pro subscribers only. There is no public API. A team that wants to try the idea has to go through the Canva app.
Canva's broader claim is harder to measure but easier to notice. A model that keeps every output editable fits a working design file. A model that returns a finished picture sits in a separate category of tool.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
