People and relationships

Find the distinct people in a video and match them against your face catalog

Run face recognition on a file downloaded with fetch_asset, which arrives with recognition off

Add another reference face (a new angle or light) to a known person to improve recognition

Name a person, or correct their name

Correct a person's name everywhere at once: shot descriptions, the Summary tab and search results update instantly, with no re-analysis

Add a named person to the catalog

One person's full record by id (p394): name, title, note, sex, how many videos they appear in, and the list of appearances

Every known person, named or not, with how often and where they appear

Ranked named people someone might actually be, for "this is someone else, who?"

See a person's faces as cropped pictures. For where a face sits in the frame, use people_sheet

See where a person's footage was filmed, by GPS

See which face sources were merged into a person, as in "Faces in this record"

Confirm the people in a video, which index_video waits for

People recognition was not sure about, strongest suggestion first

Merge two entries that are the same person

Merge many entries into one person in a single call, for someone split across many Unknowns after indexing

For each named person, gather up the Unknown cards that are really them

Reset and regroup the face catalog, splitting a wrongly merged group back into separate people

Find clips whose shots still name someone who, after recluster_identities(apply=True), is no longer in them

Move known people onto a new recognition model without re-indexing

Recreate a person's missing face picture without re-scanning the clip, for a person showing a blank silhouette

Assign an Unknown face in one clip to a known person

Hide an Unknown face in one clip (Ghost, in Review faces)

Ghosted or forgotten people who can be restored, as on the People page's Ghosts tab

Mark a detection as not a face at all (jeans, food, a blur). Ghost hides a real person; this says there was never a person

Record that two people are not the same person, permanently

Remove a person from the catalog

Mark a detection as not a person and move it to Quarantine, where it can be recovered

Undo forget_person or the app's Ghost button and bring the full record back

Bring a quarantined detection back into the people catalog, for a real face that was wrongly rejected

Detections judged not to be people (animals, statues, masks, bad crops), which can be recovered

Remove one clip's wrong sightings from a person, by the clip's video path

Remove a person tag you added to a clip, the reverse of tag_person

Tag a person as appearing in a video. person is an id (p3) or a name; a new name creates a new person

Add a title or role, notes, or sex to a person, for use in edits, lower thirds and graphics

Set a person's picture from an image file and lock it so indexing cannot replace it

Link a person to an account in another tool

Remove an account link from a person, by provider or link_id

Report clips with broken face or analysis data. Read-only by default

Who is in a video, when, and where in the frame, with each face's box [x0,y0,x1,y1] (0 to 1) in frames[].boxes[]

Whether the face engine (YuNet and SFace) is installed, how to set it up, and how many people are in the catalog

Set the minimum confidence for YuNet face detection (0.1 to 0.99)

Recheck people already in the catalog to catch faces that were never human (animals, statues, masks)

The unnamed faces waiting for a name, strongest probable match first, each with the named people it might be and why. Read-only

Compare YuNet with Apple Vision on the same frames of your own clips

Turn Apple Vision face detection on or off (off by default)

Whether Apple Vision face detection is available on your Mac and switched on

Match the faces in one indexed photo to your people catalog, as detect_people does for video

Match faces in many photos to your people catalog, a limited batch at a time

List indexed images that have not been analyzed yet. Read-only

Preview an image analysis run without running it. Read-only

Analyze photos that have not been analyzed yet, a limited batch at a time

Record how two people are related, by their pN ids, so you can later ask "who are my clients" or "who is in Zane's family"

Remove the relationships between two people: all of them, or only the type named in kind

Relationships for one person (by pN id), or all of them

Every relationship type, with how many relationships use it

Rename a relationship type

Merge two relationship types that mean the same thing, such as "client" and "My clients". Duplicates collapse to one and from_type_id is removed

Delete a relationship type

A person's full family (parents, children, siblings, grandparents, aunts and uncles, nieces and nephews, cousins), worked out from the parent, child and sibling links you recorded

Every relationship written out in plain text, grouped by person, using names where known and pN ids where not

People probably related to someone but not yet linked, ranked by family links and shared surnames. For example, a parent's spouse may be a mother

The matching relationship the other way round, using the catalog's sex field (an uncle's nephew or niece, a son's father or mother), or null when there isn't one

Create a circle: a named group, such as a workplace, whose members are connected by belonging to it. Once enough of its people have footage, it also gets a Library collection of everything they are in

Add people to a circle, many pN ids at once

Take people out of a circle, many pN ids at once

Change a circle's name, kind, member_label or note, or one member's role and dates

Every circle, or only one person's circles (person_id), with kind and member count

One circle by rcN id, with each member's name, role and dates

Merge two circles for the same place. Members join; the merged circle is removed

Delete a circle by rcN id

Suggest circles from relationships you already recorded, such as everyone linked as coworkers or classmates

Search suggestions as you type: matching people, places, cameras, events, projects and text, each with its clip count, most specific first

Where you are and what the AI models are being told. A cheap, good first call in any session

Set text added to the start of every vision model prompt (index_video, analyze_style, track_subject)

Set text added to the start of every reasoning model prompt (summaries and insights)