Make a Flat Review Clip From 360-Degree Location Footage

You recorded the room in every direction at once. The person who has to approve the location wants an ordinary video they can watch in a browser tab with the spacebar. The file doesn't change between those two sentences. The frame does, and somebody has to aim it.

Reframing is where to look from a position the camera already occupied. It is not a second camera position. The tripod stayed where it was, the doorway is still over there, the bench is still across from it. What you choose is the direction the flat view faces and how much of the surroundings it takes in. Everything else — how big the room is, what sits behind the camera, how long that walk really takes — was settled the moment the sphere was recorded, and no setting you type afterward can revise it.

Worth saying plainly at the start, because a rotated 360 view is persuasive. Pan a sphere and the room seems to slide past, and it gets easy to slip into the vocabulary of moving a camera around a set. You didn't move anything. You turned your head.

Check the sphere before you aim at it

Before choosing a direction, confirm four things about the source: which file it is, that you're allowed to use it, what projection it actually stores, and which parts of it contain usable picture. The last two are the ones people skip.

The projection matters because the rectangle you're looking at is not a photograph. It's a map. An equirectangular frame — the usual 2:1 layout, something like 5760×2880 — spreads the whole sphere across a rectangle where longitude runs left to right and latitude runs bottom to top. The horizon lands as a straight line across the middle, which is exactly why the map is so easy to mistake for an image. Toward the top and bottom, the horizontal scale swells into smears. That's the map working correctly, not a defect, and it's the signature that tells you what kind of file you have.

Not every wide file is that. A dual-fisheye recording stores two circles. A file that has passed through some other export step may carry no usable projection at all. Some cameras write their projection into the file's metadata; note what it says if it says anything, but confirm it against what you see, because metadata survives edits that changed the actual layout. Assume equirectangular input and you get a filter that does something — and the something will be wrong in a way that looks like a settings problem rather than an input problem.

Then coverage. A 360 camera sees everything except the parts it can't. The bottom of the sphere is usually where the mount, the pole, the tripod head and the stitched underside live, and it's the region where a flat view is least likely to hold up. There may be one or two seams where the lenses meet — vertical lines running the height of the map. Where a seam crosses something close to the camera, or something that moved, stitching misaligns: a person walking past it can smear or double, and a straight architectural edge sitting on it can show a step.

Finally, movement. Even a tripod shot drifts if the rig was nudged. Whatever the source camera did during your selected interval ends up inside your clip, and it interacts with framing in a specific way, because the sphere is stored in the camera's own coordinates: a fixed yaw is fixed to the camera, not to the room. If the camera turns during the shot, your flat clip turns with it. Whatever you centred slides off to one side — the bench drifts out of frame, the wall the camera swung toward arrives in the middle — and the room reads as sliding rather than sitting still. A slow deliberate pan carries less disturbance than a nudge, but it moves the frame either way, and if the question your clip answers needs the door held in view for twenty seconds, a pan spends some of those seconds pointing somewhere else. Which it is, you can only tell by watching the whole passage.

Keep an untouched copy of the source and a written copy of your settings. Two files, one decision each.

Decide what the rectangle has to explain

A flat clip isn't "the room." It's an answer to a spatial question, and the question should be chosen before any parameter is set. Otherwise you tune by taste and end up with the most attractive frame rather than the most useful one.

Make the question concrete enough to be checkable. "How does the entrance relate to the prep bench?" is something a clip can settle. "Show the kitchen" isn't — it can't be right or wrong, so it can't be aimed.

Once you have the question, name two things: what has to stay visible together in a single frame, and what you're willing to leave outside it. For the entrance-and-bench question, the doorway and the work surface have to share the frame, because their relationship is the entire point. The refrigerator behind the camera, the wall the viewer never faces, the ceiling — those can go, as long as the note that ships with the clip says what was left out.

Sometimes the honest answer is that no single rectangle will do it. If the real question is "what does the route through this room feel like," one flat view from one corner can't answer it, and widening the field until both ends are technically inside the frame won't fix it either — it will just make the far end small, stretched and unreadable. In that case, plan two clearly labeled views from the same position, or a floor plan beside the clip, or a reader-controlled tour of the location instead of a directed one. Two honest clips beat one that overclaims.

A flat view is projected, not cropped

Here's the operation people expect to perform: open the map, cut out the plausible-looking middle band, call it a view. Don't.

A crop of the equirectangular file is a slice of the map, and it keeps the map's geometry. Under some conditions it looks almost convincing, because vertical lines near the middle of an equirectangular frame stay vertical, the horizon stays level, and nothing announces itself as wrong. But the map's horizontal scale is angular — a degree of longitude occupies the same number of pixels everywhere along the equator — while a real lens looking at the same scene distributes its pixels differently, magnifying the edges of its field relative to the center. The two disagree. Push a crop away from the equator and the disagreement becomes obvious. Cut across the left and right edges of the file and you get a hard seam where the sphere's last degree meets its first.

A flat view gets projected out of the sphere rather than sliced from it, and FFmpeg's v360 filter is built for that. Its documentation describes the controls this job needs: it takes an input projection and an output projection, orients the result with yaw, pitch and roll, and sizes it with a horizontal, vertical or diagonal field of view. Here, the input is the equirectangular sphere and the output is flat, and everything else in the call is direction and coverage.

Yaw, pitch and roll are three ways to point the view, and they're easier to learn through the room than in the abstract. Yaw turns left and right, and it's the one you'll use most, because spatial questions are usually horizontal. Pitch tilts up and down, and you'll use less of it: tilting up spends frame on ceiling, tilting down spends it on floor, and the floor nearest the camera is where the sphere is usually least reliable. Roll tips the horizon diagonally. For a location review clip, roll is almost always a mistake. A level horizon is the reference the viewer uses to judge everything else, and giving it up buys nothing.

Now the limit of the tool, stated where it matters: the filter orients a view inside the sphere that was recorded. It does not move the camera. Changing yaw doesn't walk the viewer three steps toward the door; it points the same eye in a different direction. If the review genuinely needs footage from over by the door, and the camera never stood there, that footage doesn't exist and no parameter produces it. Say so in the note rather than hunting for a setting that gestures at it.

The shape of the call, as a schematic:

ffmpeg -ss 00:01:12 -i source_360.mp4 -t 20 \
  -vf "v360=input=equirect:output=flat:yaw=65:pitch=0:roll=0:h_fov=100:w=1920:h=1080" \
  -c:v libx264 -crf 18 -preset slow -pix_fmt yuv420p -movflags +faststart \
  review_door_to_bench.mp4

Two things about that block. The option names come from the filter documentation — the rolling HTML page at ffmpeg.org, checked in September 2026 — and the geometry those controls describe is what this method depends on. But no sphere was projected for this article, no installed FFmpeg build was tested, and no clip was exported. Treat the line as a shape to check against your own build rather than a command with a known result. Confirm the option names in the version you have, set the output dimensions to the frame you're actually delivering, and verify the first and last frames of your trim rather than trusting the timestamps you typed.

Two directions from one tripod

The worked scenario here is constructed — a method, not a record. Say the source is a fixed tripod shot in the middle of a small prep kitchen, with the entrance door on one side, a stainless bench across the room, and a stitch seam running vertically through the wall that holds the door. In the file's own coordinates, the door centers around yaw 25 and the bench around yaw 105. Those values belong to this example; your file will put the door wherever it puts it, and the relationship is the point, not the numbers.

Now produce two directions. The first answers the spatial question: how do you get from the entrance to the work area? Both the door and the bench have to be in the frame, and they sit about 80 degrees apart, so centering the view between them at yaw 65 gives each one roughly equal margin. Pitch stays at 0 — level, which is what the viewer's eye expects and what puts the horizon where they'll use it as a reference. Roll stays at 0. That's the relationship view, and it answers one thing well: the door is left of the bench, and the walk between them is short.

The second direction answers a different question — what's actually on the benchtop? Point the view at yaw 105, at the bench itself, and the door leaves the frame entirely. That's fine: a bench close-up was never going to explain the doorway anyway. What matters is that both clips came off the same tripod, at the same height, in the same room, and the difference between them is a direction rather than a position. If someone later asks why the second clip looks like it's standing at the bench, the answer is that it isn't standing anywhere. It's looking.

Each clip has to say what it leaves out. The relationship view doesn't show the refrigerator behind the camera, or the wall the camera never faces. That isn't a failure of the clip, it's a boundary the note should name — because a viewer who knows only what a rectangle shows will assume the rectangle is the room.

Field of view: wider is not more complete

Field of view is the second choice, and it's the one that most often goes wrong, because wider feels safer. It isn't.

Hold the editorial question constant and change only the field. At the relationship direction — yaw 65, pitch 0 — a horizontal field of about 100 degrees holds the door and the bench with roughly ten degrees of clearance on each side. Everything that matters is inside the frame, with room to spare, and nothing in the frame is idle.

Widen it to 140 and the geometry starts charging you. A flat view is a pinhole projection, so its two fields are locked to the shape of the frame: in a 16:9 output, the tangent of half the horizontal field divided by the tangent of half the vertical field equals the frame's aspect ratio. A 100-degree horizontal view works out near 68 degrees vertical. Push horizontal to 140 and vertical lands near 114 — a frame running from well above the ceiling lights to well below the camera's own height. You've added a lot of floor and a lot of ceiling, and none of it answers the question.

The extra width isn't neutral, but notice where it goes. Widening the field doesn't push the doorway toward the edge: it sits 40 degrees off the axis at 100 degrees, and 40 degrees off the axis at 140, because its direction in the room hasn't changed and only the border around it has. Its proportions come from that angle, and they are the same in both frames. What a wide frame does stretch is whatever lands on its border: this projection magnifies the edge of the field relative to the center, which is why a face at the edge of a wide-angle shot looks wider than it is. At 140 degrees in this room, the border is near floor and ceiling — magnified, and no more informative for it. That's the trade in one line: a wider field isn't a more complete explanation, it's a less precise one with more room attached.

If the relationship genuinely doesn't fit in a readable field, that's information about the location, not a reason to keep widening. Go back to two views.

Watch the whole passage, not the thumbnail

A clean still establishes nothing about a clip. Scrub it, play it, play it again.

Watch the seam, especially where it crosses something near the camera or something in motion — a person walking through it is where a stitched frame tends to come apart, and a straight architectural edge sitting on it will show any misalignment as a step. Watch the pole region if any part of your framing tilts toward the floor nearest the camera. Watch for the moment the source camera moved, if it did, and check whether your locked direction turns that into a lurch. Watch the framing itself across the entire interval: if you're cutting twenty seconds, the door has to stay in frame for twenty seconds, not just at the instant you chose the settings.

Then write down what you did. Source file identity, the interval, the direction, the field of view, the tool and its version, the output file. Without that record, the clip becomes unreproducible within a week, and if anyone asks whether the view was rotated or reframed, you'll be guessing.

And check the export where the reviewer will actually see it, not only in the application you exported from. A file that plays in your timeline can still arrive as a black rectangle on someone else's phone, and that failure has nothing to teach you about framing.

If a seam or the pole region shows up as a real defect in your chosen view, name it in the note. Don't fill it in. A 360 sphere records where the camera stood, including the parts it recorded badly, and painting over a bad seam with invented pixels produces a frame that claims coverage the source doesn't have. That's the same error as implying the camera moved, wearing different clothes.

Ship the clip with the sentence that goes with it

The deliverable is a playable file and a short note, and the note isn't decoration — it's what makes the file usable.

The note carries the question and the answer. Something like: Flat review clip from the 360 source recorded at [location]. The view faces the entrance and the prep bench together; entrance left of frame, bench right. Horizontal field 100 degrees, level horizon. Not shown: the wall behind the camera and the walk-in refrigerator, both of which the sphere captures and this frame does not. Add the interval and the settings, then point to the sphere file for anyone who wants to look around instead of looking where you aimed.

That last part matters more than it sounds. A directed clip is a good answer when the recipient has a question with a shape. When their question is open — "what's this place like" — one rectangle will keep failing, and the right move is a navigable reference alongside the clip rather than a wider and wider flat view. Similarly, if you're starting from photographs rather than recorded footage, you're building source material rather than reframing it, and the geometry you'll be managing is a different problem entirely.

So: aim the frame at the question, write down where you aimed, and hand the sphere over with the clip. If the recipient ever needs to know what was around the corner, the source still has it. The clip was never supposed to.

Frequently asked questions

What is the difference between reframing 360 footage and moving a camera?

Reframing chooses the direction and coverage of a view from a position the camera already occupied. It does not create a new camera position. The room's dimensions, what sits behind the camera, and how long a walk takes are fixed once the sphere is recorded; changing yaw or field of view cannot revise them.

How can you tell what projection a 360 source file uses?

An equirectangular frame is usually a 2:1 rectangle with the horizon as a straight line across the middle and horizontal smears near top and bottom; those smears are the map working correctly. A dual-fisheye recording stores two circles. Metadata may state the projection, but it should be checked against the visible layout, because metadata can survive edits that changed the file's actual arrangement.

Why isn't cropping the middle band of an equirectangular file the same as making a flat view?

A crop keeps the map's geometry: vertical lines near the middle may stay vertical and the horizon level, but horizontal scale is angular, while a real lens magnifies the edges of its field relative to the center. Cropping away from the equator makes the disagreement obvious, and cutting across left and right edges creates a hard seam. A flat view must be projected out of the sphere, as with FFmpeg's v360 filter.

What do yaw, pitch, and roll control, and why is roll usually avoided?

Yaw turns left and right and is used most because spatial questions are usually horizontal. Pitch tilts up and down and spends frame on ceiling or floor; the nearest floor is often the least reliable part of the sphere. Roll tips the horizon diagonally, and for a location review it is almost always a mistake because a level horizon is the reference the viewer uses.

Does a wider field of view make a flat review clip more complete?

No. Widening adds floor and ceiling and magnifies whatever lands on the frame's border; an object's direction in the room, and therefore its proportions, do not change. A wider field is a less precise explanation with more room attached. If a relationship does not fit in a readable field, the recommended move is two clearly labeled views from the same position.