Etsy allows shoppers to upload a photo — of a piece of furniture, a piece of clothing, or anything else — and find visually similar items from across its marketplace. The feature launched for iPhone users in November 2022 and uses computer vision to match the photo against approximately 100 million listing images.

The image search system converts every listing image into a mathematical representation called a vector embedding, using a type of deep learning model trained on both seller product photos and buyer review photos to handle real-world camera quality. When a shopper uploads a photo, the system finds the closest matches among approximately 100 million indexed images using an approximate nearest-neighbor algorithm. The feature was Etsy's first GPU-backed service and originated at an internal hackathon. The underlying model evolved from basic image classification, through triplet loss embeddings, to a multitask learning architecture for improved accuracy across categories.