File metadata is one layer
Image tools can store prompts, seeds, model names, workflow graphs and editing history in file metadata. This tool removes supported text and provenance containers from JPG, PNG and WebP images without changing compressed image data.
| Carrier | Cleanup in this version |
|---|---|
| PNG generation text | Removes tEXt, zTXt and iTXt chunks, including Stable Diffusion “parameters” and ComfyUI “prompt” or “workflow” fields. |
| JPG metadata | Removes APP1 EXIF/XMP, extended XMP, APP11 JUMBF/C2PA, APP13 IPTC and comments. |
| WebP metadata | Removes EXIF, XMP and C2PA chunks and updates container size and flags. |
| PDF provenance | Only common document metadata and supported XMP streams are cleaned. PDF C2PA attachments can remain. |
SynthID and pixel watermarks are not removed
Invisible watermarks encoded in pixels, including SynthID, are not file metadata. Because image data is preserved, those signals remain. Visible watermarks, clues in the image, external records and visual AI detection are also outside this tool’s scope. Removing metadata is not a guarantee that an image cannot be identified as AI-generated.
Removing content credentials has a consequence
C2PA content credentials can carry signed information about a file’s source and edits. Removing a supported container removes that attached provenance and can affect verification. It does not erase records held elsewhere or prove that the image was made by a person. Keep the original when you need its credentials.
The report identifies supported carriers rather than validating a credential’s signature or authenticity. A C2PA label means a supported carrier was detected, not that its content was verified. ICC profiles are retained by default and can contain descriptive profile information; use the optional switch if you want the entire profile removed.
Check the disclosure and provenance requirements of the service where you plan to publish. Metadata cleanup does not change your responsibility to describe the image accurately.