2026-09-06
How to cut the 20 seconds that matter out of dash cam footage (without wrecking the original)
The thing about incident footage is that you almost never have "the video." You have a memory card holding dozens of files a few minutes each, one of which contains the eleven seconds that matter, and a recording loop quietly overwriting the whole lot while you decide what to do about it. Getting those eleven seconds out is the easy half. The half that guides skip is this: the clip you hand around and the file you keep are two different objects, and confusing them is the one mistake here that is genuinely hard to undo.
The camera did not record one file. It recorded forty.
Dash cams and most security cameras do not write one long recording. They loop. Nextbase's own explainer of loop recording is the plain version: a dash cam "continuously records footage in a loop, which means that the Dash Cam records video by overwriting older video captures with new video footage," and the recording is chopped into fixed segments because "it becomes easier to manage when the recordings are segmented into smaller and more manageable files." The segment length is a setting, and it varies by model: Nextbase says "common options include 1 minute, 3 minutes and 5 minutes depending on the Dash Cam you choose," with the 522GW splitting "recordings up into three-minute segments" and the 422GW letting you "choose from one, two and three-minute clips."
Two things follow, and both shape the work. First, an hour of driving is somewhere between twelve and sixty files, so the first job is not trimming — it is finding which file the moment is in, and reading the camera's clock in the corner of the picture rather than trusting file timestamps. Second, the moment can straddle a boundary: the impact at the tail of one segment, the aftermath at the head of the next. NearVid cannot join two files — there is no merge, and saying so before you start beats discovering it at the download step. If your incident crosses a segment boundary you either produce two clips and label them, or you use a desktop tool that concatenates.
And the loop is still running. On most cameras the emergency or parking event locks a file so it is skipped by the overwrite, but the safe move is to get the relevant segments off the card, onto a computer, before you do anything else at all. No editing decision is worth losing the source to.
Before you trim anything: the original is the evidence
Here is the part that reorders the whole workflow. The National Dash Cam Safety Portal, the route UK police forces use to take footage from the public, is explicit in its FAQ about what it wants: "All that is required is a copy of the original file in unedited, uncompressed format." Not the highlight. The file. And handing it over does not discharge you of the original: "It will be your responsibility to ensure that the footage is securely saved in its original format and can be made available if required by the police." The same FAQ passes on Crown Prosecution Service advice that footage "should not be in the public domain as this may adversely affect subsequent proceedings" — aimed at posting to social media, but worth reading twice before your clip goes anywhere.
None of that makes trimming wrong. It makes trimming a second artifact. A short clip is the right thing to send an insurance adjuster who has forty claims open, to show a lawyer in the first five minutes of a call, to put in front of a fleet manager, or to look at yourself twenty times while you work out what actually happened. It is the wrong thing to be your only copy. The rule that covers every version of this — police, insurer, employer, court, neighbour — is: keep the original file, untouched, and treat every clip as disposable. This is a workflow note and not legal advice; requirements differ by country, by force, and by who is asking.
What a lossless trim keeps — and what it quietly drops
NearVid's trim is a stream copy. The video and audio packets for the window you asked for are copied into a new container without being decoded and re-encoded, which is why it finishes in milliseconds rather than minutes and why the picture is bit-for-bit the picture your camera recorded. There is no second-generation quality loss, no softened plate, no smeared brake light. That is the right shape for footage that might get scrutinised.
But a stream copy still rebuilds the container, and dash cams put things in containers that are
not tracks. On cameras built around Novatek chipsets — the Viofo A119 and a long list of relatives
— the GPS trail is stored in vendor-specific boxes rather than a standard metadata track. A
detailed
teardown of the format describes a "gps " box inside moov that "stores the file
offset (in bytes) and size (in bytes) for each GPS data box," with the actual coordinates, speed
and bearing sitting in free boxes scattered among the data chunks, each identified by
a "GPS " magic string. Those are not streams. A remux copies the streams it maps and writes a fresh
container around them, so vendor boxes like these are not something to count on surviving the
trip — check the trimmed copy in your camera's own player before you assume the map still moves,
and if the telemetry is part of your point, hand over the original.
The opposite case is the reassuring one. Some cameras never put the data in the file at all: dash2gps, a tool that reads coordinates out of Nextbase recordings, exists because "some DashCam do not store GPS location in the video file itself. They embed the location in the video frame instead," in that model's case "in the bottom left corner of the video." Burned-in coordinates, and the burned-in clock nearly every camera draws, are pixels. Pixels survive a trim, and they survive a conversion too.
Where the cut actually lands
NearVid's trim takes typed start and end times, and because a lossless cut can only begin on a keyframe, the clip may start a beat earlier than the number you entered. For most uses that is an annoyance. Here it is close to a feature: extra seconds of run-up are context, and context is exactly what a viewer who was not there needs. Type a generous window anyway — ten seconds before the moment and ten after is a better clip than a tight four-second cut that starts mid-event and invites the question of what you removed.
The re-encoding paths cut exactly where you type, because an encoder is running and can start a new keyframe wherever it likes. Convert to WebM does this everywhere, and Convert to MP4 does it in browsers whose own H.264 encoder NearVid can detect. Use them for a clip going into a group chat or onto a slide. Do not use them for the copy someone will treat as evidence: re-encoding is precisely what wording like "unedited, uncompressed" is warning against, and the frame-perfect in-point is not worth the argument.
Why this is the wrong file to hand to an upload-based converter
Consider what is actually in a two-minute dash cam segment: your own plate on the reflections, the other party's plate in full, faces on the pavement, the front of a house that is probably yours, the road you take home at the same time every day, and — since most cameras record the cabin — whatever was being said inside the car in the moments after an accident. That is a dense little package of personal data about several people, at least one of whom has not agreed to anything.
Every browser-tab converter that "supports large files" works by uploading yours, which means a copy on a server you do not control, retained under a policy you did not read, at the exact moment you have not yet decided whether to share the footage at all. NearVid runs a real FFmpeg build compiled to WebAssembly inside the page, so the file is opened by your own browser and no bytes leave the machine — the file picker is the only transfer that happens. Deciding, after watching your own clip, not to send it to anyone is a completely normal outcome, and it only stays a private decision if the trimming step never involved a stranger's server.
Worth saying plainly: your camera's own software may already do this locally, and if it does, use it. Garmin's manual for the Dash Cam 46/56/66W describes trimming inside the Garmin Drive phone app — "drag the handles on the video progress bar left or right to trim the video length," then Export. That is a local edit too. NearVid is for the common case where the desktop software is Windows-only, the phone app will not open a file already copied to a laptop, or the footage came off a camera whose vendor app was abandoned three years ago.
The order to do it in
- On the camera, lock or protect the relevant recording so the loop stops eyeing it. Then copy the whole segment — and the one before and after it — off the card onto a computer.
- Put the originals somewhere you will not touch again, and write down the file names plus the time shown by the camera's own clock in the picture.
- Work on a copy. Drop it into NearVid; the file is stashed locally and the workspace opens.
- Trim a generous window and download. Rename the result so nobody can mistake it for the original — something with "clip" in the name.
- Send the clip. Keep the original, and say you have it. If a form asks for the unedited file, give it the unedited file.
If you also need one sharp frame — a plate, a signal, a road sign — the frame extraction card pulls the exact timestamp you type as a JPG or PNG at the recording's own resolution, without re-encoding the video; the still-frame post covers choosing a timestamp that is not a motion-blurred mess.
What NearVid cannot do here
The list matters more than usual for this job, so it goes in full. There is no merging, so an incident spanning two segments becomes two clips. There is no blurring or redaction — no way to obscure a bystander's face or an uninvolved plate, which for anything you plan to publish rather than submit is a real gap, and a reason to use something else. There is no stabilisation, no rotation, no annotation, no arrows, no subtitle track, and no way to strip the cabin audio out of a trim; a lossless cut copies the audio track exactly as recorded, so if the conversation in the car is the problem, that clip is not shareable in this form. There is no "make it fit under X MB" target either, only three fixed quality presets on the re-encode paths and a labelled size estimate beside them.
Size is a hard boundary rather than a slow path: NearVid processes the whole file in memory, so it takes up to 500MB per file on a desktop and 50MB on phones and low-memory devices. A one-minute 1080p segment is comfortably inside both; a three- or five-minute 4K segment will often clear the phone number on the very phone you are holding, which is another reason the workflow above moves the files to a computer first. And nothing here can recover what the loop already overwrote, or add any kind of authentication: a trim is a convenience, not a chain of custody.
The honest summary
Loop recording means your footage arrives as many short files that the camera is actively overwriting, so the first move is to get the segments off the card, not to edit them. The portal that UK police use asks for "a copy of the original file in unedited, uncompressed format" and puts the duty of keeping that original on you, which makes a trimmed clip a second artifact for humans in a hurry rather than a replacement. NearVid's trim is a true stream copy — no re-encode, no quality loss, milliseconds — and it will start on the nearest keyframe, so type a generous window. Burned-in clocks and coordinates survive it because they are pixels; vendor GPS boxes are not streams and may not. It cannot merge, blur, stabilise or mute, and it stops at 500MB on a desktop and 50MB on a phone. What it does do is run entirely inside your browser, which is the right shape for footage of an incident you have not yet decided how to handle.