Back to home

A CLOSER LOOK

There’s more to a file than the picture.

A plain-English guide to AI metadata: what we read, what we remove, and what stays outside the scope of a scan.

A record of how a file was made

Some creative tools attach provenance metadata when they create or edit media. This can include C2PA Content Credentials or an IPTC digital source type inside XMP. These records are stored in the file container, separately from the encoded picture or video.

Why a real photograph can carry AI information

A generative edit can add a source declaration to an otherwise camera-captured photo. That declaration describes tool use; it is not a universal judgment about whether the whole image is synthetic.

Meta describes metadata indicators as part of its labeling approach. TikTok also uses Content Credentials and other signals. A file scan cannot reproduce either platform’s entire system.

What the scan tells you

We identify supported metadata blocks, extract source declarations where possible, and show the limits of the platform evidence. We do not authenticate the creator, verify C2PA certificate chains, or use the absence of metadata as proof that a photo is real.

What cleaning removes

The cleaner removes supported C2PA, XMP, EXIF and legacy IPTC blocks. Those blocks can also contain dates, location, orientation and copyright information. Keep your original if you need that information.

The engine does not re-encode the image or video data. Metadata-dependent display behavior, such as EXIF orientation, can change. Review the downloaded file before sharing it.

What “verified” means here

We scan the output again and check that the requested supported metadata is gone and no recognized AI declaration remains. A SHA-256 checksum identifies the exact output bytes. Verification is about that file operation; it is not a promise about a platform’s future labeling decision.

Supported photos and videos

Supported containers include JPEG, PNG, WebP, TIFF/DNG, MP4 and MOV. The limit is 200 MB per file. Unsupported or incompletely parsed files are refused.

What this tool does not remove

Visible logos, captions burned into video, and watermarks embedded in pixels such as SynthID are outside the tool’s scope. We do not detect or verify their removal. We also cannot remove a label already attached to a published post.

Review your result

Review the scan findings, remove supported metadata, and download the verified result. Keep your original file so you can compare the output.

Read the platform guidance

Cleaning metadata does not change a platform’s disclosure rules. See Meta’s labeling approach, TikTok’s creator guidance, and TikTok’s additional detection and watermarking signals.

QUICK ANSWERS

A few more details.

Try a free scan
What is an AI label watermark?

People often use this phrase for AI provenance metadata: information attached to a file about how it was created or edited. It is different from a visible logo or an invisible watermark embedded in pixels. This tool works on the metadata layer.

Does removing it change my photo?

The engine does not re-encode the image or video data. It removes supported metadata blocks, including EXIF and XMP that may contain orientation, dates, location or copyright information. Metadata-dependent display behavior can change, so keep your original and check the result.

What does the scan check?

The scan checks supported metadata blocks for AI source declarations and other file information. It shows what was found before you choose to remove it.

Does it work for video?

Supported MP4 and MOV containers can be scanned and cleaned. The current limit is 200 MB per file. Files we cannot fully parse are refused rather than returned as verified. Video labels can also depend on signals outside file metadata.

Do I need an account?

Not to scan — that is free and unlimited. Cleaning requires registration and a paid plan.

Does this guarantee no AI label?

No. We verify removal of supported metadata, not the outcome of a platform’s labeling system. Other signals, self-disclosure or a later edit can still lead to a label. This also cannot change labels on posts you have already published.

Can you remove a watermark hidden inside the picture itself?

No. Pixel watermarks such as SynthID are outside this tool’s scope. We don’t detect, remove or verify them, and a clean metadata scan says nothing about whether one is present.

Create your account

Sign in to your Labelwatermark account.

or continue with email
How metadata cleaning works · Labelwatermark