Canva launched its proprietary vision-language model, Brush, on October 29, 2026. This tool is designed to assist users with design tasks by interpreting visual layouts, layers, and styles. The model is capable of performing both minor adjustments and full presentation redesigns based on written instructions.
The model was developed by Canva Research, a 160-person internal AI laboratory. Brush operates by analyzing the live design canvas and refining its output in real time. A key feature is that all edits remain fully editable by the user, ensuring continued control over text, imagery, and overall layout structure.
Canva reports that Brush increases design editing speed by up to 69% compared to previous AI agents. Furthermore, the company estimates that the model reduces the cost of editing tasks by 46%. These improvements are part of a broader strategy to transition from third-party AI providers to proprietary systems.
The training process for Brush utilized anonymized user data and real-world design tasks. This development follows the recent release of Canva Craft, an image generation model from the same team. The company aims to integrate these in-house tools to better align AI capabilities with specific design workflows.
