Visual search finds images by what they look like rather than by their tags or filenames. Instead of typing words, you start from an image — one already in the library, or one you provide — and the system returns assets that are visually similar: same subject, same composition, same style.

It works through vision embeddings: each image is converted into a numerical representation of its visual content, and similarity becomes a matter of distance between vectors. This makes visual search independent of metadata quality — an untagged photo is just as findable by appearance as a perfectly described one, which is exactly what keyword search cannot offer.

The everyday uses are concrete. A designer has one shot from a shoot and wants the rest of the series. A marketer found the perfect hero image but needs a portrait-orientation alternative in the same style. An editor suspects a picture already exists in the library before requesting a new one. In each case, describing what you want in words is harder than pointing at an example — visual search closes that gap. It also pairs naturally with duplicate detection: similarity search finds things that look alike, while perceptual hashing flags things that are effectively the same file.

In freedam, any image asset offers a "find similar" search backed by vision embeddings, so visually related assets surface even when their metadata has nothing in common.

Related terms: Semantic Search, Duplicate Detection