Guide · Pillar

How to Search Video Footage: Seven Ways to Find the Shot in Your Head

You already know something about the shot you want. This guide turns whatever you remember, a line of dialogue, a face, a colour, a feeling, into a query that finds it.

4 min read

Every search starts with a fragment: a line somebody said, a face, a colour, the fact that it was the wide one from the second day.

Converting that into a filename means guessing at the shorthand you used at 2am. FrameQuery searches from the fragment. There are seven ways in.

What you rememberWhat to doExample
The exact words somebody saidQuote them"we should push the launch"
Roughly what was said, not the wordsDescribe it in plain languagetalking about delaying the launch
Who is in itName them@sarah
What is physically in shotName the thingwhiteboard
What the shot feels likeDescribe the mood and lightquiet empty office at dusk
How it was framed or shotUse a filtershot:wide angle:eye-level
Only the file nameType it as-isA001_C012
Nothing, but you have one good shotOpen it and use Find Similarno typing at all
Start from what you already know, not from what the tool can do

Search what was said

Every indexed clip has a transcript with sentence-level timestamps, and each result carries the timecode of the line that matched. Clicking it takes you there.

Type the words plainly and you get clips containing all of them, anywhere in the clip, in any order. Put quotes around them and you only get clips where those words appear together, in that order.

You want

The bit where the CEO said the launch was slipping

Type

"push the launch"

Without quotes you also get every clip that mentions a launch and separately mentions pushing. With quotes you get the sentence.

You want

Any discussion of the budget, no matter how it was phrased

Type

budget

Unquoted single words are the right tool when you want breadth. Start here, then narrow.

You want

The soundbite, but only from the interview, not the panel

Type

"we had to start over" -panel

The exclusion strips the coverage you already know you do not want.

Quotes are the difference between browsing and finding

Quote when you are sure of the wording. Do not quote when you are sure of the topic. If you are not certain of the exact words, use the next section.

Search what was meant

You remember the gist of what someone said, not the sentence. Quoting fails here, and keywords often do too, because people rarely use the word you are thinking of.

Describe it instead. talking about leaving her job finds a clip where someone said "I handed in my notice in March", with no words in common.

You want

Someone explaining why the project nearly failed

Type

explaining what went wrong with the project

Matches the sense of a passage. The speaker never has to use the word "failed".

You want

The moment somebody thanks their family

Type

thanking their parents

Works on "I owe all of this to my mum and dad" with no shared vocabulary at all.

You want

Anything where a person sounds unsure about a decision

Type

hesitant about making a decision

Tone and stance. Keyword tools cannot do this one.

Describe the substance. Do not try to guess the words.

A useful habit

Write the query as if you were describing the clip to an assistant editor over the phone. Full sentences carry more meaning than a keyword list.

Search who is in it

Name a person once and they become searchable everywhere. Type @ and the name, and you get every clip where they appear on screen or speak.

A search for @sarah also returns the clip where Sarah is talking off camera while someone else is in frame. Face-only tools miss those.

You want

Everything with Sarah in it

Type

@sarah

On screen or on mic. Use quotes for a full name: @"Sarah Chen".

You want

Sarah talking about the merger specifically

Type

@sarah "the merger"

The person narrows it to one speaker, the phrase to one moment.

You want

The two of them in the same shot

Type

@sarah @james

Two-hander coverage in one query.

You want

James, but not the town hall footage

Type

@james -town-hall

Recurring events dominate a person's results. Subtract the one you are not after.

The @ operator narrows any other search to one person

Naming happens in Manager → People. Enable recognition, then assign matching faces and voices to a person. Voice cards show a source filename and a waveform you can scrub; click the filename to check the same interval in the source clip before assigning. You can create the first person directly from Assign, or search an unnamed voice with Search videos in its actions. You can also right-click a speaker tag in a clip's transcript panel to name whoever is talking. The voice preview steps show the full flow.

The deep dive on finding people across your footage walks through that setup and what to do when somebody's results look thin.

Search what is on screen

Objects are detected per scene, so you can search for a thing rather than a description of a thing. whiteboard, bicycle, laptop, dog. This covers footage with no dialogue: drone plates, B-roll, timelapses, cutaways. "Do we have anything with a coffee cup in it" is one search.

You want

Cutaways to use over a voiceover about the office

Type

laptop keyboard desk

Several object terms broaden the net. Any of them appearing is a hit.

You want

Product shots without anyone holding it

Type

bottle -hands

Excluding an object controls the composition of the results.

You want

Every shot with a vehicle in it, from a mixed library

Type

car van truck

Objects work on footage with no transcript.

Objects are literal. Name the thing.

Search the feel of a shot

You can search the frame itself with a description of what it looks like, and get matches where no description or transcript exists.

empty street at dawn. harsh fluorescent office lighting. crowd shot, hands in the air. quiet and cold. Images are matched directly against the words.

You want

An establisher that feels lonely

Type

empty landscape under grey sky

Light, space and weather are in the frame and in no transcript.

You want

Something warm to cut to after a hard scene

Type

golden hour, backlit, people laughing

Colour temperature plus an action narrows further than either alone.

You want

The claustrophobic interview setup, not the airy one

Type

tight framing against a dark wall

Framing plus environment.

Mood searches. Vague is fine here, as long as it is visual.

Short for browsing, long for finding

A single feeling word works. Type peaceful and you get the calmer end of your library. When you need a specific shot, lengthen it: still water at sunrise, no people.

The deep dive on searching by look, mood and framing covers colour, shot type and camera angle.

Search how it was shot

Sometimes you need the 4K material, or the ProRes, or everything from the second camera, or only the slow motion.

Filters use field:value and stack freely with everything else.

FieldValuesWhat it is for
codec:prores, braw, r3d, h264, hevc, dnxhrPull only the format you can cut natively today.
res:or resolution:4k, 1080p, 8kFind the deliverable-grade material in a mixed library.
camera:any make or model in the file metadataSeparate A-cam from the crash cam or the drone.
fps:24, 25, 50, 120The fastest way to isolate slow-motion coverage.
container:mov, mp4, mxfUseful when the same shot exists as both original and proxy.
lens:any lens recorded by the cameraMatch coverage shot on the same glass.
iso:800, 3200Track down the shots that will be noisy before you grade them.
date:a shoot dateScope to a single day of a multi-day shoot.
tc:or timecode:a timecode valueJump to a logged moment from a paper edit.
shot:or shot_type:wide, medium-wide, close-up, extreme-close-upAsk for framing rather than content. Hyphenated, though spaces and underscores are accepted.
angle:or shot_angle:eye-level and other hyphenated anglesFind the cutaway that matches your master's eyeline.
Filter fields. Type them as field:value, or use quotes for multi-word values.

You want

Only the slow-motion coverage from the A camera

Type

fps:120 camera:FX9

Two filters and no search term is a valid query.

You want

Wide shots to open the sequence, in a cuttable format

Type

shot:wide codec:prores

Framing plus format.

You want

The low-angle hero shots of the building

Type

building shot:wide angle:eye-level

Filters narrow, the plain term supplies the subject. Mix them freely.

You want

Deliverable-grade material only, from one shoot day

Type

res:4k date:2026-05-14

Only the material you are allowed to deliver.

Filters combine with each other and with plain search terms

Combine, then subtract

All of the above compose. Go wide, look at the results, then subtract. Adding terms narrows on content. Subtracting terms removes the category of result that keeps showing up.

"exact phrase"

Exact phrase

Only matches when those words appear together, in that order.

"final cut"

@name

Person

Narrows to a person you have named, whether they are on screen or speaking.

@sarah or @"Sarah Chen"

-word

Exclude

Removes results containing that word. Works on phrases too.

-rehearsal or -"work in progress"

field:value

Filter

Restricts by a technical or shot property. Combine as many as you like.

codec:prores shot:wide

plain words

Meaning

No operator at all. Describe what you are after and every layer is searched at once.

someone looking nervous before going on stage

The full query grammar. Everything else is plain description.

Worked example. You want the moment your founder talks about the early days, from an interview, not from a conference stage, in wide framing so it cuts against the close-ups you already have:

@maya "when we started" shot:wide -conference -stage

A person, an exact phrase, a framing filter and two exclusions in one line.

Read the badge

Every result carries a badge saying which layer matched it.

transcriptWhat was said

The words in the spoken audio, timestamped to the moment they were said.

e.g. we should push the launch to Q3

sceneWhat was happening

A written description of each scene, so you can search editorial context rather than keywords.

e.g. two people shaking hands in a lobby

visualWhat the frame looks like

The image itself, matched against your words directly. No description has to exist first.

e.g. moody neon alleyway at night

objectWhat is in shot

Individual things detected in the frame, listed per scene.

e.g. whiteboard

metadataHow it was shot

Camera, codec, resolution, frame rate, lens, ISO, timecode and date, read from the file.

e.g. camera:FX9 res:4k

filenameWhat it is called

The literal file name, for when a DIT hands you a card number and nothing else.

e.g. A001_C012

Every result carries a badge naming the layer that found it

If your results are all filename, your query is only matching file names, so the description is not landing. If they are all transcript when you wanted visuals, you are describing what people say instead of what the camera sees.

When a search comes back wrong

Empty or useless result sets have a short list of causes.

Nothing at all, for a phrase you are certain about

Quotes demand an exact, in-order match, and transcripts vary on filler words and contractions. Drop the quotes and let the meaning match do the work.

Trypush the launch

Nothing at all, for footage you know you have

You are probably scoped to one project. The toggle beside the search bar switches between This project and All projects whenever a project is selected.

Trysame query, set to All projects

A person returns far fewer clips than expected

Their faces or voices are sitting in more than one identity. Open the People tab, assign the leftover identities to them, or use Merge With to fold a duplicate person in.

Try@sarah, after merging

Results are technically right but all from one shoot

One dominant event is crowding everything else out. Subtract it and see what is underneath.

Tryinterview -"town hall"

Every result is a filename match

Your terms are not matching content. Rewrite the query as a description of what happens in the shot rather than a label for it.

Trytwo people talking across a desk

A mood search is in the right mood but too broad

A single feeling word is a browsing query. To narrow it, add what the shot physically contains.

Tryperson alone at a window, low light

A filter silently kills the query

Filter values must exist in the footage. If no clip is tagged close-up, shot:close-up returns nothing. Remove the filter and check the results carry that property at all.

Tryclose on hands

When a search comes back wrong, one of these is almost always why

When you cannot describe it at all

When no query gets close, find one result that is near and use Find similar scenes on it. It matches the shot instead of your words. You get scenes that share its content and its look. There is a Find similar lines alongside it for transcripts, so you can take one good soundbite and pull every other moment where somebody said something like it.

Use it when a search gets to roughly the right place.

Build the habit

Three habits.

Name your people early. A few minutes in the People tab on a new project makes @ work for the entire life of that footage.

Search descriptions, not labels. Type what a stranger would see, not the shorthand you would write on a folder.

Subtract more than you add. When the results are wrong rather than empty, exclusion usually fixes it faster than another term.

The table at the top of this guide covers which of the seven to start from.

Common questions

What is the best way to search video footage when you only half remember the shot?
Describe it in plain language rather than guessing at keywords. FrameQuery searches spoken words, scene descriptions, detected objects and the frames themselves at the same time, so a sentence like 'someone laughing at a kitchen table' can match footage that was never tagged or transcribed with any of those words.
Can you search video by what someone said?
Yes. Every indexed clip is transcribed with sentence-level timestamps, so each result carries the timecode of the line that matched and you land on the moment the phrase was spoken rather than the top of the file. Put the words in quotes for an exact phrase, or describe the gist without quotes to match the meaning instead.
How do you find every clip a particular person appears in?
Add the person once in the People tab and assign their detected faces and voices to them, then search @theirname. FrameQuery matches both faces on screen and voices in the audio, so the results include clips where they speak off camera as well as clips where they are visible.
Why does a video search return no results?
Usually one of four reasons: the footage has not finished indexing, the query is in quotes and the exact phrase never occurs, an excluded term is removing more than you expect, or you are scoped to one project instead of the whole library. Widen the scope first, then drop the quotes.
Can you search video without an internet connection?
Yes. Indexing needs a connection, but once a clip is indexed its search data lives on your machine. Searching runs locally with no network round trip and no per-query cost.

Try it on your own footage

Index a folder and search it the way this guide describes. Searching your library is free and works offline once indexing is done.