natural-language-image-search

Why we included this project

If you have ever wanted to see what CLIP actually does with images and text, this project is one of the cleanest ways to try it. It pre-computes embeddings for about two million photos from the Unsplash dataset, then lets you search them by typing a phrase like "two dogs playing in the snow" and returns the nearest matches by vector similarity. The included Colab notebook walks you through the whole pipeline, from downloading the data to building the index to running queries, so you can follow along without having to reimplement the hard parts. Teams that want semantic search over their own image collections can reuse the feature-vector approach directly, and there is also a lighter path that filters Unsplash's Search API results through CLIP when you don't want to host the full dataset.