How our AI-Powered Search Works

TL;DR — Our system doesn’t require us to send your photos or search queries to a third party. Everything happens right in our own system, preserving your privacy while delivering powerful results.

At a high level, semantic image search is like having an assistant who understands both language and image content equally well, allowing you to search for photos using a descriptive phrase instead of just keywords. It recognizes what you mean and finds images that match that meaning. This is an immense step forward in the field of image search, and it’s powered by advances in machine learning.

Each photo in our archive is processed by a machine learning model, essentially a pre-trained ‘brain’ that has seen and learned from millions of images and associated text. When the model encounters an image, it generates a mathematical representation or ‘embedding’ of that image. An embedding is like a fingerprint of an image that can be used to represent it in a way that a computer can understand and compare.

When you type in a search term or phrase, that happens again — this time with your phrase. The same model processes this phrase to generate a corresponding embedding. This is crucial for our system, as it allows it to understand your search in the same terms as the images it has to search through.

When you make a search, our system takes the fingerprint of your query and compares it to all the fingerprints of the images in our database. This isn’t a simple ‘yes’ or ‘no’ comparison but a measure of how close your search is to each image, on a scale from identical to totally different. This process of measuring closeness is called cosine similarity, and it’s what lets us rank the images in our database in terms of relevance to your search.

What makes this process really stand out is the use of pre-trained machine learning models. These models have been trained on a vast amount of data, and they don’t require us to send your photos or search queries to a third party. Everything happens right in our own system, preserving your privacy while delivering powerful results.