CLIP is a fascinating model released by OpenAI—back in January 2021, when they were still doing most things in the open—that can embed both text and images.

Crucially, it embeds them both into the same vector space.

If you embed the string “dog”, you’ll get a location in 512 dimensional space (depending on your CLIP configuration).

If you embed a photograph of a dog, you’ll get a location in that same space… and it will be close in terms of distance to the location of the string “dog”!

This means we can search for related images using text, and search for related text using images.

I built an interactive demo to help explain how this works. The demo is an Observable notebook that runs the CLIP model directly in the browser.1


  1. Embeddings: what they are and why they matter, Simon Willson