Image replacement in Canva designs using reverse image search

Canva implemented a scalable similar-image replacement system using reverse image search to maintain their high-quality image library. This involved selecting a model for image embeddings, experimenting with five high-performing models, and ultimately choosing DINOv2 for its superior results in preserving image subjects and background. The new system significantly speeds up image replacement in templates and is now integrated into Canva's Template Assistant, allowing for human review and quality control. Future improvements may include better handling of symbolic and text-heavy images.