AI Metadata Remover for Video
Strip C2PA manifests, XMP packets and provenance tags from Sora, Veo, Runway and Kling exports. The video streams are copied bit for bit, so quality is untouched.
Drop your videos here
MP4, MOV, M4V, WebM or MKV - nothing is uploaded
No videos yet. Drop a file above to see what it carries.
| File | Before | After |
|---|
Important: this removes metadata, not watermarks. Invisible pixel watermarks such as Google SynthID are encoded into the picture itself and survive a remux. Anything written into the file's metadata - C2PA manifests, XMP, creation tags - is removed.
No Re-encoding
The audio and video streams are copied unchanged and only the container is rebuilt. Your quality is identical and a long clip takes seconds instead of minutes.
Nothing Is Uploaded
The whole remux runs on WebAssembly inside your browser. Your footage never reaches a server, is never stored and no account is needed.
Shows What It Found
Before and after are listed side by side, so you can see exactly which C2PA manifests, XMP packets or generator tags were in the file and that they are gone.
Handles the Big Four
MP4, MOV, M4V and WebM cover what Sora, Veo, Runway, Kling, Luma and Pika export, plus anything straight off a phone or camera.
What AI Video Generators Write Into Your File
Every major AI video tool now signs its exports. Sora, Veo, Runway, Kling and Firefly write C2PA Content Credentials into the container, and most add XMP packets and creation tags on top. Those manifests state which model produced the clip, when, and sometimes which prompts or edits were involved. Upload the file anywhere and that record travels with it, which is exactly what the standard was designed to do - and exactly what you may not want when the clip is a private draft, a client deliverable or a piece of internal research.
Where the Data Actually Sits
In an MP4 or MOV file the C2PA manifest lives in a top-level uuid box with a fixed identifier, XMP sits in a second uuid box, and free-form tags end up in udta. WebM keeps the same information in Matroska tag elements. None of that is part of the picture - it is bookkeeping wrapped around the video and audio streams, which is why it can be removed without touching a single frame.
Remuxing Instead of Re-encoding
This tool runs FFmpeg compiled to WebAssembly with -map_metadata -1 -map_chapters -1 -c copy. The streams are copied byte for byte into a freshly built container, so nothing is decoded and nothing is re-compressed. Boxes the muxer does not recognise - including the C2PA uuid box - are simply not carried over. The result is visually identical to the original, produced in seconds even for a large file, and the before-and-after scan on this page proves the metadata is gone.
What This Cannot Do
Metadata is not the only provenance signal. Google's SynthID and similar systems embed an imperceptible pattern into the pixels and the audio waveform, and that pattern survives remuxing, re-encoding, cropping and screen recording. This tool makes no attempt to defeat it and no honest tool can claim to. If your goal is to understand what a file discloses, run it through the Content Credentials checker first; if your goal is to remove the metadata layer, that is exactly what happens here.