How to Remove Metadata from Video: GPS Tracks & Device Names

Everyone has heard that photos carry EXIF data. Far fewer people know that a video file carries more, in more places, and that the most revealing part of it is not metadata in the usual sense at all: it is a separate track inside the file that logs where the camera was, several times a second, for the entire recording.

Here is what is actually inside the video files people share every day, where each piece hides, and what it takes to remove it without ruining the video.

A video is a container, and everything writes into it

An MP4 or MOV file is not one blob of video. It is a tree of boxes: one holding the compressed picture, one the sound, and a crowd of small ones describing them. Every tool that touches the file — the phone, the camera, the editor, the export preset, the uploader — writes into that crowd. Nothing ever cleans up after itself.

• Container tags: title, artist, author, copyright, comment, description, album and genre, often filled in automatically by an editor from your project settings.
• Encoder and software fields naming the exact application and version that produced the file, which is a precise fingerprint of your workflow.
• Creation and modification timestamps, stored twice: once as a readable tag and once inside the mvhd, tkhd and mdhd headers, where most tools never look and almost none of them clean.
• Handler names on each track. A phone writes something like 'Core Media Video'; an Android device has been known to write 'ISO Media file produced by Google Inc.'; an action camera writes its own model name.
• A vendor code in every sample description, four bytes identifying the manufacturer — or, after a careless cleaning pass, identifying the tool that cleaned it.

The part that actually matters: telemetry tracks

Phones and action cameras record more than picture and sound. Alongside them they write a timed-metadata track, a third stream that carries a stream of sensor readings with timestamps attached. GoPro calls its format GPMF. Apple uses a track type called mebx. Android phones write CAMM. Sony writes RTMD.

What is in them varies by device, but GPS is common: latitude, longitude, altitude and speed, sampled continuously. A single photo leaks one location. A ten-minute clip with a telemetry track leaks a route — where you started, the way you went, how fast, and where you stopped.

The reason this survives so often is mundane. A telemetry track is not a tag, so the usual instruction to clear a file's metadata does not touch it. Worse, the standard way to copy a video into a new container copies every track it finds, so a tool that strips the tags and remuxes the file will faithfully carry the GPS log into the 'cleaned' output. The only fix is to decide, deliberately, not to copy that track.

AI video and Content Credentials

AI video tools have adopted the same provenance standard as AI image tools. A C2PA manifest, also called Content Credentials, is a signed block naming the service that generated the clip, the software agent involved, a creation timestamp and a declaration that the source is trained algorithmic media. In a video it usually sits in a uuid box, which is a general-purpose container that tools are free to ignore — and most do, which means they leave it in place.

Content Credentials exist for good reasons, and we go through the trade-offs in our guide to C2PA. The point here is the same as for images: it is there, it names your tools and your timing, and you should know about it before you publish.

Why re-encoding is the wrong way to remove it

The obvious way to strip a video clean is to open it and export a new one. It works, and it costs you the video. Re-encoding decodes every frame and compresses it again, and a second round of lossy compression is visible: softer motion, blocking in dark scenes and gradients, a duller image at the same file size. Do it twice and it compounds. It is also slow, turning a two-second job into several minutes for a long clip.

There is no need for any of it. The compressed video and audio can be copied into a fresh container without being decoded at all — the same packets, byte for byte, in a new wrapper. That is what a container rewrite does, and it is why cleaning a video should take seconds rather than minutes and cost you nothing in quality.

What has to be kept

Removing metadata should never change how a video plays, and a couple of things look like metadata but are not.

• The rotation flag. Phones record in landscape and tag the file with a rotation for players to apply. Strip it and every portrait clip plays on its side. It says nothing about you, so it should be read before cleaning and written back afterwards.
• Track language codes and timing information, which describe the file rather than its author.
• Subtitle tracks, which are part of the content you meant to share.

A checklist before you publish a video

1. Scan the file first, so you know whether any of this applies to your particular clip. A screen recording and a holiday video have very different problems.
2. Pay attention to whether a telemetry or data track is listed. That is the one that leaks a route rather than a point.
3. Clean without re-encoding, so you are not paying for privacy with quality.
4. Check that the result still plays the right way up and still has its audio, and that the duration matches.
5. Keep your original. The metadata-rich file is useful in your own archive for proving authorship or reconstructing a shoot.

You can scan and clean a video with our free video metadata cleaner. It lists every field it finds before removing anything, drops telemetry tracks rather than copying them, and checksums the cleaned file against your original so you can see that every frame survived.