1. What do we inspect inside the file?
We first parse metadata that can be read from the file, including EXIF, XMP, selected GPS fields, and camera or editing-software information when available.
We also inspect the file structure and text for recognizable C2PA, JUMBF and Content Credentials markers, plus certain strings associated with modern generation or editing tools such as OpenAI, Gemini and ComfyUI.
- EXIF and readable camera or device information.
- GPS coordinates when they can be parsed from the file.
- XMP and common descriptive fields.
- Recognizable C2PA, JUMBF and Content Credentials signals.
- Textual AI-tool or workflow signals when they are actually present in the file.
2. How does cleanup work?
Cleanup does not modify the source file. The browser decodes the visible image into pixels and creates a new file from those pixels. Ordinary metadata from the source is not intentionally copied into the new output.
The output is then inspected again with the same workflow so you can compare readable fields and file signals before and after cleanup.
3. What happens to file size and format?
The current release also tries to reduce file size. Non-transparent images may be exported as JPEG at a balanced quality setting, while images that need transparency use WebP or PNG where appropriate and supported.
For that reason, we do not describe the process as lossless. The goal is a smaller, cleaner sharing copy with good visual quality while the source remains unchanged.
4. What do we not claim to remove?
Metadata cleanup is not the same as removing every possible provenance signal. Invisible marks embedded in pixel data, including some watermarking systems, are a separate layer from EXIF or XMP.
External platforms can also use their own signals and policies. MediaClean therefore does not promise that cleaning file metadata will change a label or classification decision made by X, Instagram or another platform.
5. Privacy in the current processing path
Core inspection and cleanup for supported images run in the browser. The source file does not need to be uploaded to a MediaClean server for these current functions.
If future features require server-side processing, they will be separated from the local workflow and disclosed before a user chooses to use them.
6. How do we verify the output?
We do not stop after generating a new file. The output is scanned again and compared by readable metadata fields, recognizable signals, file size and output type. That gives users something concrete to review instead of a generic success message.
We also use careful terminology: detecting a marker means the file contains a signal worth inspecting, not that a complete cryptographic provenance chain has been verified unless the tool explicitly performs that validation.