Semantic search finds footage by meaning, not by exact words: "skeptical at first" finds "I honestly didn't think this would work" and the visual of a raised eyebrow. It combines what's said, what's shown and who's on camera, and returns the segment playing from the right second.
Semantic search finds footage by what it means, not by the exact words. "Skeptical at first" returns "I honestly didn't think this would work", "I was not convinced", and the shot where the creator raises an eyebrow before the reveal.
It works because every segment is turned into a representation of its meaning: the transcript, the AI's description of what's on screen, and the video itself. Your search is turned into the same kind of representation and the closest segments come back, ranked. Verbatim search is the exact-words complement for claims and scripted lines.
The practical effect is that you search the way you'd brief an editor. "Creator applying serum, morning light, says it absorbs fast" is one query with three parts, and the library answers all three.
Let's say a strategist types "the moment they realise it works". No creator said those words. The search returns 9 reaction segments across 6 creators, because the meaning matches even though the sentences don't.
Both. The AI watches the video and describes each moment, and the video itself is indexed, so "dropper close-up on a clean background" works even in a clip with no talking.
Drive searches file names and, at best, a document's text. It does not know what happens at minute 4 of a video. Semantic search works at the moment level, inside the video.