How to Find Every Clip a Person Appears In

Person search connects detected faces and voices to names you choose. Once those links exist, use an @name query to find matching footage, including moments when somebody speaks off camera. Select a project to keep the search within one job, or choose All projects for the wider library.

Turn biometric analysis on first

Face and voice detection is opt-in. Until you enable it, a fresh library has no people and @ has nothing to match.

Enable recognition in Manager → People. Setup explains the local models, the data stored and the consent choices. Let enabled analysis finish before expecting faces and voices to appear. For missing detections in a selected project, Library tools → Rescan project can analyse its footage; source media and the required models must be available.

Name people once, search forever

Once analysis has run, FrameQuery groups faces and voices across your library into identities. Naming them happens in the People tab.

The flow is short:

1. Use Add Person to create a name, or start with Assign on a voice card and create the person there.
2. Work through Faces and Voices, assigning each matching identity to that person.
3. Search @thatname.

The name applies to linked matches across your indexed library, including footage from other shoots. Search results still respect your project selection and other filters. Recognition can miss or split a match, so check the source when an identity is uncertain.

Preview a voice before assigning it

The Voices section now leads with source filenames rather than generated voice IDs. Each card shows a preview interval and a waveform drawn from that preview's audio.

1. Find a voice by its source filename. Use Search source videos, sorting and pagination to work through a larger collection.
2. Press Play or drag the waveform to a useful point. You can pause and resume; scrubbing while paused selects a position without starting audio. The arrow keys also move the focused timeline.
3. Click the filename to open the source clip at that spoken interval. The recording gives you more context than a voice ID can.
4. Choose Assign and select the person, or create a new person. Different voices belonging to the same person can share that name.
5. Use Search videos from the card's actions. An unnamed voice can be searched directly; a named person can also be searched with @"Sarah Chen".

Voices 2

Interactive illustration · no audio

interview-take-04.mp4

0:42–0:47

0:00 / 0:05

8 appearancesAssign

location-interview-02.mp4

1:18–1:23

0:00 / 0:05

3 appearancesAssign

A voice you can check before you name it.

Drag a waveform or try Assign with the example person, Sarah Chen.

Search videos

A preview needs an accessible original clip or usable local proxy. If it fails, check the source and retry. A missing preview does not mean that the underlying voice has no indexed matches. Likewise, a similar-sounding preview is a reason to check the source, not a guarantee of identity.

Speaker diarization and voice recognition solve different problems. Diarization separates speakers within a transcript. Voice recognition links matching voices across recordings; assigning those voices to a person makes them searchable by name. For exact spoken words, combine the person with dialogue search.

Two shortcuts:

• Naming a speaker while you watch. Open a clip, and in the transcript panel right-click the speaker tag beside any line to assign or reassign it to a person. You are hearing them as you name them.
• Jumping straight to their footage. Open a person or voice card's actions and choose Search videos. The search starts with the name or detected voice already selected.

On screen and on mic

@sarah returns clips where Sarah is visible, and clips where Sarah is speaking, including the ones where she is not in frame at all.

A face-only search misses:

• The interviewer asking the question from off camera
• Sarah's reaction line while the shot holds on someone else
• Anything shot over her shoulder
• The director talking to camera operators between takes
• A voice on a speakerphone in the middle of a meeting

Combine a person with everything else

On its own, a person search returns matching footage within the current search scope. The useful queries pair @ with one more constraint.

You want

The exact thing Maya said about the funding round

Type

@maya "the funding round"

Person plus exact phrase. Lands on the second she said it, not the top of the file.

You want

Maya talking about money, however she phrased it

Type

@maya talking about raising money

Person plus meaning. Use this when you remember the subject and not the sentence.

You want

Usable close-ups of her for the cutaway

Type

@maya shot:close-up

Person plus framing.

You want

Her, in 4K, so the shot is deliverable

Type

@maya res:4k

Keeps matches at 4K resolution and filters out lower-resolution footage.

You want

Anything with her that is not from the weekly standup

Type

@maya -standup

One recurring meeting will otherwise dominate every person search you run.

You want

The two of them together

Type

@maya @dan

Both present. Two-hander coverage in one query.

Names with spaces go in quotes: @"Maya Fernandez". Without them the search only takes the first word.

Fixing thin results

If someone's results look too small, first check the selected project and active filters. Next, search an unassigned voice directly from its card. If that finds the missing footage, assign the voice to the person you expected to find. Their faces and voices may be spread across more than one identity.

They get separated by real things: a haircut, glasses on and off, a hat, a dramatic lighting change between a lit interview and a handheld corridor shot, or a camera angle far from the others. Voices separate for the same sort of reason, mostly recording conditions.

There are two fixes in the People tab, depending on what you find:

• The extra identity is unassigned. Assign it to the existing person, the same way you named them in the first place.
• You accidentally created two people. Use Merge With on one of them. Every identity moves onto the person you keep, and the duplicate is removed.

A person's card shows how many identities are linked to them. A low count on a person who appears often means some identities are still unassigned.

A person returns clips from one shoot and nothing from another

Check the project filter first. If both shoots are in scope, look in Faces and Voices for another matching identity and assign it to the person you already created.

Try@maya, after assigning the second identity

The same person exists twice in Known People

Use Merge With on the duplicate. Its linked identities move across and the duplicate is deleted.

Try@maya, after merging

@fullname returns nothing but @firstname works

An unquoted mention stops at the first word, so @Maya Fernandez searches for the person Maya plus the separate term Fernandez.

Try@"Maya Fernandez"

Nobody appears in the People tab at all

Check sign-in, consent and model setup, then let analysis finish. For missing detections in a selected project, use Library tools → Rescan project and check that its source media is accessible.

Tryenable analysis, then @maya

Results include the right person but useless coverage

Person search is broad by design. Add a framing filter or subtract the recurring event that is crowding the list.

Try@maya shot:medium-wide -standup

Practical workflows

Pulling one person's selects. Search @person, add the framing you need, then work down the results by score.

Assembling an interview. Search @person plus a description of the topic for each beat of your paper edit. You get candidate soundbites per beat, with timecodes.

Reviewing a person's footage. Choose All projects and search @person to review detected matches across the library. Check unassigned identities and source clips when coverage looks incomplete.

Finding the conversation, not the person. @a @b plus a topic finds the exchange.

A note on what is stored

Biometric analysis is off until you turn it on, and the People tab tells you what it stores before you enable it. Names stay on your device. Searching for a person publishes nothing.

If people-related data matters for your work, whether that is client footage, minors, or anything covered by biometric privacy rules, read the privacy detail on people matching before you build a workflow on top of it.

Where to go next

The main search guide covers the other six ways in. If most of what you are looking for is spoken rather than seen, searching dialogue is the more useful next read.