What metadata can reveal
AI-generated images may carry C2PA or Content Credentials containers, digital-source fields, software names, and PNG text chunks containing prompts, seeds, samplers, or model settings. These records are different from visible watermarks: they are data structures attached to the file and may remain even when the image itself looks ordinary.
How browser-based cleaning works
The AI Metadata Remover scans for supported C2PA markers, PNG text chunks, and provenance-related EXIF or XMP fields. It then re-encodes the visible pixels and checks the new file again. The optional fingerprint setting makes subtle pixel changes so the exported file has a different SHA-256 hash; all processing stays in the browser.
How to remove the metadata
Inspect the image before cleaning so you know which provenance fields were actually found. Remove the supported metadata, wait for the verification pass, and download the new copy. Leave the optional pixel-change setting off when you only want metadata removal and want to minimize visual or compression changes.
- Inspect: Read recognized metadata locally and group it by privacy relevance.
- Remove: Re-encode supported photos or rewrite supported PDF metadata structures.
- Verify: Parse the output again before enabling a verified success state.
Verification and important limits
A metadata cleaner cannot remove an invisible pixel-level watermark such as SynthID, erase records already stored by a platform, or guarantee that visual classifiers and similarity search will not recognize an image. A changed file hash only proves that the bytes changed. It is not proof of anonymity or detection evasion.
Choose the right cleanup workflow
Use the Photo / EXIF tool when the main risk is GPS, camera, author, or capture-date information. Use this AI workflow for file-level provenance and generator parameters. Keep disclosure, copyright, and platform rules in mind before publishing a modified file.