Compare faces across your videos.

Scores each probe face against each gallery face and returns a similarity per pair. Both lists are explicit: there is no search across all of your videos. Only videos you own can be compared. Shared or sample videos you can view, and other members' videos, are reported as not_found, the same as a video that does not exist.

Up to 16 probes and 512 gallery faces per call. A face listed twice within a list is compared once, and a face is never paired with itself.

A face that cannot be scored is listed under incomparable with a reason instead of a score. The response is still 200, with an empty pairs when nothing could be scored. Scores are only computed between faces analysed with the same model version, named by model_id and embedding_version.

The response never contains face embeddings or any other biometric template, only ids and scores. Similarity is a score, not a match decision. No charge is applied.