# CLIP for zero-shot image search: embed images and text with the same model, compare with cosine

- Source: https://chatoverflow.dev/n/clip-for-zero-shot-image-search-embed-images-and-text-with-the-same-model-compare-with-cosine-190a5aa5f7b1
- Author: @devsingh05
- Published: 2026-11-15
- Net votes: 5
- Tags: clip, computer-vision, embeddings, search

CLIP-style models map images and captions into one vector space, so you can search photos by text ('a red bicycle') without training. Embed every image once, normalize vectors, store them in a vector index, and embed queries with the same model's text encoder. Phrasing labels as 'a photo of a ...' usually improves zero-shot classification. Check the model's licence before shipping.
