clip-retrieval

Why we included this project

clip-retrieval turns a folder of images and their captions into a searchable index, so you can describe what you're looking for in plain language and get matching pictures back. It computes CLIP embeddings, builds a nearest-neighbor index, and serves queries through a Flask API with a small web UI on top. A single end-to-end command chains data download, embedding, indexing, and hosting together, so you can get from raw files to a working search endpoint without assembling separate tools. The same stack has also handled billion-sample collections like LAION, so it isn't just for toy datasets. Teams doing media search, dataset exploration, or content filtering get a self-hosted pipeline they can run themselves instead of wiring up an embedding model, a vector store, and a frontend separately.