TL;DR: Stripping metadata from a video online without uploading is possible because modern browsers can parse and rewrite MP4/MOV/WebM container boxes locally in JavaScript. The cleaner walks the ISO Base Media File Format tree, deletes the udta and meta atoms at file and track level, removes EXIF blocks from any embedded thumbnail, and strips the C2PA manifest from the JUMBF box if present. The video bitstream (H.264, HEVC, VP9, AV1) and audio samples (AAC, Opus) are passed through byte-identical — frames, codec, bitrate, length unchanged. Output is typically a few KB smaller than the input. The whole pass runs in roughly 45 seconds for a one-minute clip, finishes inside the tab, and never sends bytes to a server. Honest limit: stripping the file closes the metadata read at upload; it does not remove audible audio watermarks or pixel-level visual marks the AI tool may have baked into the frames.
Search "remove metadata from video online" and the top results all share a feature that should make you suspicious: they want you to upload your video to their server. For a tool whose entire purpose is privacy, that's a structural contradiction.
The point of stripping metadata is that you don't want certain things about the file to travel. The first thing those tools do is take the file somewhere outside your control.
There's a different way. The browser can strip metadata from a video file entirely on your own device — same tab, no upload, no server, no log of what got cleaned anywhere off your computer. This post is the workflow, the mechanism, and what the strip actually does and doesn't fix.
Server-Upload Tools Are Not Privacy Tools
Most online metadata removers were built before browsers had the capabilities to do real file processing locally. The architecture they're stuck with is straightforward and wrong-shaped:
- You upload the file to their server.
- Their server runs a metadata-stripping pass on it.
- You download the cleaned file back.
In that flow, your file exists in three places: your computer, the network wire (during upload), and the server's disk. Privacy policies say the server deletes the file "after processing," but you have no way to verify that. Logs persist. CDN caches persist. Backups persist. The file metadata you wanted to strip has now been read by a third party — possibly logged, possibly retained for ML training, possibly subject to subpoena. The exact set of things you were trying to prevent.
Browser-only tools work differently. The processing happens in JavaScript inside your browser tab. The file moves between your filesystem and your browser's memory, never the network. When you close the tab, the file is gone from any system other than your own. This is the architecture Metadata Cleaner uses, and it's the architecture you should look for in any privacy-claiming tool. The broader case for keeping your file off third-party servers is laid out at the EFF privacy hub.
What's Inside a Video File's Metadata
Before stripping, it's worth knowing what's actually in a video file's metadata block. Modern video files (MP4, MOV, WebM, and the ISO Base Media File Format family they all descend from) are container formats organized as nested box trees. Inside that tree, several places carry metadata:
udta(User Data) box at file level — encoder name, software version, copyright string, sometimes GPS coordinates, sometimes C2PA assertionsudtaper video track — track-level metadata: orientation, language, sometimes the source app's identifier and versionudtaper audio track — codec specifics, encoder identification, sometimes ID3-style tags carried over from a source music filemetabox at file or track level — iTunes-style key/value metadata: title, artist, comment, often used by export pipelines from CapCut, Final Cut, and Premiere- EXIF in embedded thumbnail — many video players generate a thumbnail frame on encode and that thumbnail can carry full EXIF (GPS, camera serial number, capture timestamp, lens model)
- C2PA manifest in JUMBF box — for AI-generated video (Runway, Pika, Sora, HeyGen, CapCut AI), a cryptographically signed assertion lives here (explainer here)
A video shot on an iPhone has all of this except the C2PA assertion. A video generated in Runway, Pika, or Sora has all of this plus the C2PA assertion. A video edited in CapCut with AI features (auto-captions, generative effects) has the source file's metadata plus CapCut's added assertions. The platform-by-platform breakdown of what each of TikTok, Instagram, and YouTube reads from this tree at upload is in the platform detection piece.
Stripping metadata from a video means removing all of these in one pass — file-level, track-level, embedded thumbnails, and any C2PA assertions. Anything less leaves a signal behind.
Photo by Pramod Tiwari on Pexels.
The Browser-Only Workflow
The whole operation runs as JavaScript in your tab. The file's bytes pass through your browser's memory and back to your filesystem; they never touch a server.
- Open Metadata Cleaner in any modern browser. No login. No account.
- Drag your video file into the drop zone. Supported formats: MP4, MOV, WebM.
- Click Clean. The browser walks the file's box tree, removes file-level and track-level metadata containers, neutralizes the embedded thumbnail's EXIF, and strips C2PA manifests cleanly.
- Click Download. You get back a video with byte-identical visual and audio content. Same frames, same audio, same length, same codec, same bitrate. The wrapper around them is empty.
Under the hood, the cleaner reads the file as an ArrayBuffer, walks the box tree with a small ISO BMFF parser, rewrites the boxes that need cleaning, and serializes the result back out via Blob and URL.createObjectURL. The same APIs ship in Chrome, Firefox, Safari, and Edge, including on mobile. A one-minute 4K clip finishes in roughly 45 seconds; the pass scales linearly with file size. There is no upload, no progress bar tied to network throughput, no server round trip.
For batch — clipping a long-form video into multiple shorts, or cleaning a folder of footage from a shoot — Pro is $4.99/mo for unlimited batch and ZIP downloads. The processing model is the same: every file stays in the browser, the only difference is the queue UX and the ZIP packaging at the end.
What Gets Stripped (Image, Video, and Audio Tracks)
A video file has multiple tracks (visual, audio, sometimes subtitles, sometimes data). Each carries its own metadata. The clean removes everything in one pass:
- Visual track: encoder strings, codec metadata, capture device info, GPS coordinates, color profile (replaced with the codec's default rather than the device's signed profile)
- Audio track: codec metadata, encoder strings, ID3-style tags, sometimes embedded artist or album info inherited from a music source file
- Embedded thumbnails: the small JPEG that many video files carry for player previews. This often holds full EXIF including GPS and camera serial. Cleaned along with the rest.
- File-level boxes: the
udtabox at the file root, themetabox, any custom container metadata an editor or AI tool wrote during export - C2PA assertions: removed from the JUMBF box if present, including signed manifest and validation thumbprint
- Subtitle/data tracks: usually preserved (they contain content, not metadata), but any metadata about those tracks — encoder, language tags, authoring app — is normalized
The audio inside the video is byte-identical before and after. The visual frames are byte-identical. The container around them is leaner. Players read the audio and visual streams; they don't depend on the metadata content. A cleaned file plays in QuickTime, VLC, Apple's media stack, every browser's <video> element, and every editing app.
Honest limit on what stripping the file does not do: an audio watermark baked into the audio waveform (some AI music tools and ElevenLabs voice outputs use these) survives metadata removal entirely, because it's inside the bitstream, not in the wrapper. A visible pixel watermark some image generators burn into frames also survives. Acoustic fingerprinting at the platform level (TikTok, Instagram, YouTube all run it on uploads) also persists — stripping closes the metadata read, not the classifier read. The full difference between metadata and watermark is in the watermark vs metadata breakdown.
Photo by Hanna Pad on Pexels.
Verifying the Strip Worked
Verification is part of the workflow specifically because metadata removal is invisible to the eye and creators reasonably want a check. Drag the cleaned file to metadata2go.com and compare it to the original. The cleaned version should show:
- Empty
udtacontent at file and track level - No GPS, no device serial, no encoder identification beyond generic codec descriptors
- No C2PA assertion (the section is absent, not just blanked)
- A minimal embedded thumbnail with no EXIF
- The
metabox either removed or reduced to required structural keys
If anything is still showing up, that's a bug — email and it'll get prioritized. The same check can be done locally with exiftool -a -G1 yourfile.mp4 if you prefer a terminal-based reader. Both surfaces read the same container structure; both should agree the metadata is gone.
A small caveat on third-party readers: some surface inferred metadata even when none is in the file. "Codec: H.264, Color space: BT.709" is structural — it describes the bitstream, not the source. That stays after a strip and is not a privacy signal. The things that matter (encoder strings, device IDs, GPS, C2PA) should be gone.
Photo by Juan Gomez on Pexels.
When to Strip Before Cutting vs. After
A common workflow question: do you strip metadata on the source file before editing, or strip the final export after editing?
Strip the final export. Reasoning: editing software (CapCut, Final Cut, Premiere, DaVinci Resolve) all add their own metadata to exports. If you strip a source clip and then edit it, the export from your editor will have the editor's metadata baked in. You'll need to strip again. Wasted work.
The exception: if your source clip contains GPS or device data you don't want the editor to read — some editors send telemetry that includes file metadata — strip the source first. Then strip the export again. That's a paranoid workflow but a defensible one for any project where the source footage carries a location signal you don't want stored anywhere outside your device.
For most creators, the once-after-export workflow is sufficient. Edit normally, strip the final export, upload the strip.
FAQ
Why is doing this in the browser better than a desktop app?
The browser is the lowest-friction privacy boundary that already exists on every device. No install, no permissions dialog, no auto-update channel that could change behavior without you noticing. Processing runs in the tab's sandbox; when the tab closes, every trace of the operation closes with it.
Does this work on my phone?
Yes. Mobile Safari, Chrome on Android, and Firefox mobile all support the file APIs used. Drag-and-drop becomes tap-to-pick on mobile. Larger files (a few hundred MB) may hit memory limits on older devices — a memory ceiling, not a bug. Use desktop in that case, or split the file first.
What about HEVC/H.265 video?
Supported. The tool processes the file's container structure regardless of which codec the video stream uses. H.264, H.265 (HEVC), VP9, AV1 — all handled at the container level. The bitstream itself is never decoded; only the box tree around it is rewritten.
Can I strip metadata from a video without changing the file size?
The cleaned file is slightly smaller than the original because the metadata bytes are removed. The difference is usually minimal — a few KB on a typical video, more if the source carried a large C2PA manifest with embedded thumbnails. Visual and audio quality is unchanged.
Does stripping break the video for any players?
No. Video players read the audio and visual streams; they don't depend on the metadata content. A cleaned file plays in QuickTime, VLC, web video tags, and every editing app. The only thing missing is the metadata sidebar entries in the player UI.
What if my video has chapters or interactive metadata?
Chapter markers and interactive metadata (popular in podcast video and audiobook video) live in the metadata tree. The clean removes these along with everything else. If you need to preserve chapters, add them back after cleaning in your authoring tool — or, if the chapters are non-negotiable, accept that this workflow isn't a fit and use a tool that supports selective stripping.
Will the platforms still know it's AI-generated even after I strip?
The metadata strip closes the C2PA file read, which is the strongest, fastest signal platforms use. Tier-2 classifiers (visual and audio model detection) and Tier-3 account-level signals (your past upload history, follower-graph patterns) still apply. Strip-the-file is necessary but not always sufficient on platforms with strong Tier-2 or Tier-3 enforcement.
Strip in your tab. Don't hand your file to a server to be told it's clean. Try Metadata Cleaner free and clean MP4, MOV, or WebM in the browser, with the file never leaving your device.