Faces

Tessera can recognize and group every appearance of a person across your whole library. The feature is off by default and stays dark until you explicitly opt in.

Why it is opt-in

Face vectors are biometric data under GDPR Article 9 and laws like BIPA. Tessera treats them accordingly: the entire faces lane is gated behind a single faces.enabled switch that defaults to false. Nothing about who is in your photos is computed or stored until you turn it on.

When enabled, detection and embedding run entirely on-device — there is no cloud face database and no network call. Face vectors are stored locally in a dedicated face-vector store.

Enabling it

Faces is enabled in your configuration:

faces:
  enabled: false   # set to true to opt in — biometric data is opt-in
  detector: apple_vision
  embedder: sface   # Apache-2.0, 128-dim, commercial-safe default

You can also flip it for a single run with the MP_FACES_ENABLED environment variable, without editing the config file. See Configuration for the full block.

The default embedder is SFace (Apache-2.0) — commercial-safe and the recommended choice. An alternative embedder is available for non-commercial use.

How grouping works

Once enabled, Tessera detects faces, embeds each one, and clusters them into people. Name a person once and Tessera surfaces every appearance of them across the library. Clustering uses a chaining-resistant algorithm so well-separated faces do not collapse into one mega-cluster.

Right to erasure

Every person — and all of their face vectors — is fully erasable. Deleting a person removes their biometric data from the store. Because nothing is uploaded, local deletion is complete deletion; there is no server copy to chase.