Visual clustering (body-signature)
Group photos by whole-body appearance — works for helmets, masks, distance.
Face detection breaks down for helmeted racers, masked surgeons, athletes shot from behind. Visual clustering doesn't need a face — it uses a whole-body CLIP embedding to group photos of the same person by appearance.
Enable per album in the album edit modal. Like face detection, it requires a separate Art. 9 legal-basis confirmation since the body embedder may capture face features within a body crop.
The clusters appear under Album → People → Liveries / Identifiers / Faces tabs. Promoting a body cluster to a named athlete uses the same flow as face clusters.