Guide
AI watermark vs metadata
Images touched by AI tools can carry two very different kinds of mark: metadata stored in the file, and watermarks drawn into the pixels. Only one of them can be removed by a metadata cleaner, and knowing the difference sets the right expectation.
How to use it
- 1Check the file for supported AI metadata such as C2PA or workflow text.
- 2If present, rebuild the image to remove the supported metadata.
- 3Accept that pixel-level watermarks and platform classifiers are out of scope.
- 4Verify the cleaned file's metadata with a before and after report.
Two kinds of AI mark
The terms are often used interchangeably, but they behave differently.
- File metadata: C2PA Content Credentials, XMP and workflow text stored in the container.
- Pixel-level watermarks: patterns such as SynthID drawn into the image data itself.
- Platform detection: a classifier that decides from the picture or the account, not the file.
What a metadata cleaner removes
A rebuild replaces the container, so file-based marks disappear. The pixels are re-encoded but not rewritten to remove a watermark.
| Mark | Stored in | Removed by rebuilding? |
|---|---|---|
| C2PA / Content Credentials | File container | Yes |
| XMP / workflow text | File container | Yes |
| SynthID and pixel watermarks | Image pixels | No |
| Platform classifier | Platform side | No |
Why the distinction matters
A promise to remove every AI trace is false. A precise promise, removing supported file-based indicators and being honest about pixels, is one you can rely on and verify.
Rebuilding and file bytes
Rebuilding from pixels produces a new encoding, so the exported file's bytes differ from the original. MetaRase does not measure or claim any effect on reverse image search or perceptual hashing.
How to verify
MetaRase shows a before and after report for the file's metadata. If the after-scan is clean, the supported file-based marks are gone; the pixel and platform layers remain outside the tool's reach.
Frequently asked questions
Can MetaRase remove SynthID?
No. SynthID and similar watermarks are part of the pixel data, not file metadata.
What AI marks can be removed?
Supported file-based indicators such as C2PA Content Credentials, XMP and workflow text.
Does rebuilding stop reverse image search?
Rebuilding re-encodes the image, so the exported bytes differ from the original. MetaRase does not measure or claim any effect on reverse image search.
Will platforms still label a cleaned image?
Possibly. A platform's own detection is separate from file metadata and is unaffected.
Is my image uploaded?
No. Everything runs locally in your browser.
Is this honest about limits?
Yes. The tool removes supported file-based marks and clearly states what it cannot remove.
Sources
Inspect, clean, and verify in your browser
Your image bytes never leave your device. MetaRase rebuilds the image locally, then scans the output again before you download it.
Open the cleanerRelated pages
Results describe supported embedded indicators only and cannot guarantee how a platform will classify an image.