People
Faces are detected in every photo as it is analyzed, and faces that appear to be the same person are grouped together. Naming a group once makes every photo of that person findable by name, including photos uploaded later.
Media → People.
People is marked beta. Recognition is good but not perfect, and it is designed around you confirming its guesses rather than trusting them blindly.
Name a person
The People screen shows every group of faces the system has found. Groups without a name show as Unnamed with a photo count.
Open one, confirm the faces are all the same person, and give it a name. From then on that name works as a filter in the library, a search term, and a way for an agent to find the right photo.
Two filters make the first pass quick:
- Unnamed: only people who have not been named yet
- Needs review: only people with suggested matches waiting
Review suggested matches
When the system finds a face that probably belongs to someone you have already named, it proposes the match rather than assigning it. Proposals appear as Review matches with a count.
Confirm or reject each one. Every answer improves future matching, so time spent here early pays off more than the same time later. The review panel also shows the AI's own assessment of each suggestion (likely a match or hard to tell), so you can move quickly through the obvious ones.
Correct a mistake
| Rename | The name was wrong or has changed. |
| Merge | Two groups are the same person. Pick which one to merge into. Common when someone appears with and without a beard, or across a decade. |
| Delete | Remove the person. The faces are unassigned, not deleted. |
| Change cover photo | Pick a better representative face. |
| Remove from People on a single face | This particular detection is wrong. The photo stops counting toward that person. |
| Link to person on a single face | Assign a face by hand. Manually linked faces are always kept. |
Low-quality faces (small, blurred, or turned away) are hidden by default and can be shown with a toggle. They are the most common source of bad matches.
Reindex a person
Reindex recomputes a person's auto-matched photos from scratch, using the current settings and everything you have confirmed since. Manually tagged faces are always kept; automatic matches are cleared and proposed again.
Use it after a large review session, or after merging two groups. On a large library it takes a few minutes and reports what changed: added, removed, and left to review.
Search by face
Search by Face Photo takes a photo containing a face (one of your own, not necessarily one in the library) and finds matching photos. Use it when you need every shot of someone and do not know whether they have been named yet.
Find photos of a person
Anywhere in the library, filter by People. The item screen also lists the people in each photo, and Show faces draws boxes around the detections so you can see what was found and correct it.
Consent
Recognition runs on your own library, and the groupings and names never leave your workspace. Whether to build a named index of people's faces is a question about your organization's policies and the law where you operate, not about the software. If you are unsure, name only the people who have agreed to it. The feature works with some groups left unnamed.