Video Search
Video Embeddings 101
Embeddings are the underlying technique that makes semantic video search possible. Here's what they are, in plain terms, and why they matter for editing.
The plain-terms version
An embedding is a way of turning something — a frame, a sentence, a clip — into a list of numbers that captures its meaning, not its exact words or pixels. Two things that mean similar things end up with similar numbers, even if they look or sound completely different on the surface.
That's what makes a search like "the moment someone opens a gift" work even if no clip is labeled that way: the system compares the meaning of your query to the meaning of each moment in the footage, not the literal text.
Why this beats keyword or tag-based search
Tags require someone to predict every way a moment might later be searched for, and label it that way in advance — which doesn't scale past a small library. Embeddings don't require prediction: the meaning is captured once, when the footage is indexed, and any later query gets compared against it, however it's phrased.
Where this fits into an AI editor
Embeddings are the mechanism, not the product — they're what sits underneath a search bar or an agent's decision-making, not something you interact with directly. When Broll indexes your raw footage, it's building this kind of representation for every frame, which is what lets both you and the editing agent find and use the right moment later.