Guide · Deep dive
How to Search Footage by Look, Mood and Framing
The search nobody expects to work. Describe what a shot looks and feels like and get matches from footage that was never tagged, described or transcribed.
Every other kind of search needs somebody to have said something, written something or named something. Visual search does not. You describe what a shot looks like and the frames answer directly.
That makes it the only tool that works on the half of most libraries nobody ever logged: drone plates, timelapses, cutaways, establishers, silent coverage, the second camera.
Short queries browse, long queries find
A single feeling word works. Type peaceful and you get the calmer end of your library. Type sad and the results lean that way too. This is genuinely useful when you are looking for something rather than for one specific thing, which is most of what B-roll selection actually is.
What a short query cannot do is be precise. When you need a particular shot, the fix is to lengthen rather than to rephrase.
To browse
To narrow
sad
person sitting alone looking down, low light
corporate
glass office building, people in meeting rooms
energetic
crowd with hands in the air, bright stage lights
peaceful
still water at sunrise, no people
tense
two people facing each other across a table
Both columns are valid queries. The left is how you browse a feeling. The right is how you ask for a shot. Reach for whichever matches the question you are actually asking.
You want
An opener that feels isolated
Type
empty landscape under grey overcast sky
Space, weather and light are all carried by the image and written down by nobody.
You want
Something warm to cut to after a hard beat
Type
golden hour backlit, people laughing outdoors
Colour temperature plus an action gets much closer than either on its own.
You want
The claustrophobic version of the interview setup
Type
tight framing against a dark wall
Framing plus environment. This is the search that replaces scrubbing dailies.
You want
Night exteriors with a bit of atmosphere
Type
wet street at night reflecting neon signs
Surfaces and light sources are strong visual anchors. Use them.
Lengthen, do not rephrase
When a visual search is in the right territory but too broad, add detail: light, time of day, location, what people are doing, what is in the background. This is the opposite of a quoted dialogue search, where you shorten to recover.
Objects, when you want the literal thing
Objects are detected per scene, so whiteboard, bicycle, coffee cup and dog are searchable directly. Use objects when the requirement is specific and physical, and descriptions when the requirement is about the whole frame.
The distinction in practice: laptop finds shots containing a laptop. someone working late alone at a laptop finds a mood, and will rank a lonely shot above a bright open-plan one. Both are useful. They answer different questions.
Objects are also how you control composition by subtraction. bottle -hands gets you product shots without a presenter in them, which is a genuinely tedious thing to do by hand.
Filter by how it was shot
When the specification matters more than the content, use filters. Shot type and angle turn a content library into a shot library.
You want
Wides to open a sequence
Type
shot:wide
A filter with no search term at all is a valid query. Sometimes it is the whole job.
You want
Hero shots of the building
Type
building shot:wide angle:eye-level
The filters set the grammar of the shot, the plain term supplies the subject.
You want
Cutaways that will actually cut against your master
Type
shot:close-up angle:eye-level
Matching framing and eyeline is what makes coverage usable rather than merely available.
You want
Slow motion only, from the deliverable camera
Type
fps:120 res:4k
Two technical filters. Removes every phone clip and proxy in one line.
The filter panel beside the search bar covers the things that are awkward to type: dominant colour, video duration and a date range. Colour is the one worth knowing about, because it is the fastest way to pull footage that will cut together tonally. Find the warm material for a sequence, or every shot with the cool grey-blue of your office location, without describing either.
Reach for the panel when you are steering by feel, and the query syntax when you already know exactly what you want.
When you cannot put it into words
Some shots resist description entirely. You know it when you see it and every query you try comes back with something adjacent.
Get close, then stop typing. Find one result in the right neighbourhood and use Find similar scenes on it. That matches the shot itself, so you get scenes sharing its content and its look without you having to name either.
This is the single most reliable way to build out coverage. Find one good establisher, pull everything like it, and you have your options in front of you in two moves.
A feeling word gets the right mood but too many clips
It is working, it is just a browsing query. Keep the word and add what such a shot physically contains.
Trypeaceful still water at sunrise, no people
Results are the right subject with the wrong feel
You described the content and not the treatment. Add light, time of day and framing.
Tryoffice at night, single desk lamp, empty
A shot filter returns nothing
Filter values only match footage that carries that property. If nothing is tagged close-up, the filter empties the result set on its own. Remove it and describe the framing instead.
Trytight on hands
Everything comes back as a filename match
Your terms are being read as labels rather than descriptions. Rewrite as a sentence about what is visible.
Trywide shot of a harbour at sunset
You keep getting the same handful of clips
Your query is too narrow for the library. Broaden the description and use exclusions to steer instead.
Tryharbour boats water -timelapse
Workflows this unlocks
Building a B-roll shortlist. Run three or four mood searches for the sequence you are cutting, take the best from each, and you have a selects bin without having opened a single file blind.
Matching coverage across shoots. Find the shot you are cutting to, run Find similar scenes, and pull equivalents from other shoots in the same library. Continuity of look becomes searchable rather than remembered.
Mining an archive nobody logged. Old footage with no notes is exactly where visual search has the biggest edge, because there is no metadata to fall back on and nothing else can see inside it at all.
Answering a client question in a meeting. "Do we have anything with the old branding on it" is a query, not a project.
Where to go next
The main search guide covers all seven ways in, including dialogue and people. If the footage you are working with is mostly interviews, finding a line of dialogue will serve you better than this page.
Common questions
- Can you search video footage by what it looks like?
- Yes. FrameQuery matches your description against the frames themselves, so a query like 'empty street at dawn' returns shots that match visually even when no transcript, filename or written description mentions any of those words.
- How do you search footage that has no dialogue?
- Silent footage is covered by three layers that do not need audio: detected objects, written scene descriptions, and direct visual matching of the frame. Drone plates, timelapses and cutaways are all searchable by what is in them and what they look like.
- Can you search video footage by mood or feeling?
- Yes. A single feeling word such as 'peaceful' or 'sad' returns footage that leans that way, which is useful for browsing a library rather than hunting one clip. To narrow it down, keep the feeling and add what the shot physically contains, for example 'peaceful still water at sunrise, no people'.
- How do you find shots that match coverage you already have?
- Use Find similar scenes on a result you already like. It matches the shot rather than your words, returning scenes that share its content and its look, which is faster than trying to describe a frame you can already see. There is a Find similar lines equivalent for transcript matches.
- What filters can you apply to a video search in FrameQuery?
- Typed filters use field:value syntax and cover codec, resolution, camera, frame rate, container, lens, ISO, date, timecode, shot type and camera angle. The filter panel beside the search bar adds dominant colour, video duration and a date range.
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.