captions embeddings

[sujjeee]

hey! just general curiosity. what if we create an image caption, convert it into vectors, and store it in a vector db? then, when someone uploads an image, we create a caption, convert it into a vector, and perform a similarity search to find similar images.

isn't that a good solution for image similarity search?