Text-to-image search retrieves visually relevant images for natural-language queries by placing text and image embeddings in the same vector space. A jointly trained multimodal model such as CLIP makes related text and images close together, enabling vector search over indexed image embeddings. section 15.3.2
Movie image search
A query such as “singing in the rain” searches movie-image embeddings; related query wording can produce nuanced differences, such as “superhero flying” versus “superheroes flying.”
