POST /v1/search finds timestamped moments in your video library. Send a text description, a reference image, or a public image URL: results are ranked by relevance and grouped by video.
Input types
At least one input is required. You cannot send both
image and image_url in the same request. A text-only query also requires collection_id: image search can run across your whole library, but text search needs a collection to scope against.
What gets searched
Every video is indexed across up to four signal types, and a query can match any of them:
Each segment’s
type field identifies the matching signal. See Search results for the response shape.
Scoping and pagination
Passcollection_id to limit results to one collection and reduce unrelated matches. Results are paginated with page and per_page, with a maximum page size of 50.
Credits and limits
Search charges depend on the query type. Free accounts have a daily search quota. See Usage and billing for current pricing and Rate limits for request limits.Text Search
Natural language queries against visual and spoken content.
Image Search
Find visually similar frames from a reference image.
Transcript Search
Search only what was said, not what’s shown.
Combined Queries
How modes merge, and the
modes/labels parameters.