Keyword search compared to semantic search
Keyword search on video uses titles, tags, and sometimes the transcript. It finds a word only when somebody says it or types it.
Semantic search uses embeddings of the visual content, the speech, and the on-screen text. The query and the video go into the same vector space, so the search matches meaning.
What a result looks like
A semantic video search result is a moment, not a file: a video ID, a start time, an end time, and a confidence score.
See the Video Search API.