AIwork4me/image-metadata-cleaner

image-metadata-cleaner cleans privacy-sensitive metadata from user-owned images by writing sanitized copies. The skill is designed for legitimate privacy hygiene, file preparation, and reproducible publishing workflows.

It is not designed or documented for hiding authorship, evading provenance checks, bypassing AI labels, or misrepresenting an image's origin.

What it does

• Re-encodes image pixels into a fresh output file.
• Writes copies instead of modifying originals in place.
• Defaults folder output to metadata-cleaned/.
• Refuses output paths that resolve to the same file as the input.
• Produces a human-readable summary and optional JSON manifest.
• Reopens outputs and scans for common metadata keys and provenance marker strings.

Install

Copy this folder to your Claude Code skills directory:

# Global, all projects
cp -r image-metadata-cleaner ~/.claude/skills/

# Or project-level only
cp -r image-metadata-cleaner your-project/.claude/skills/

The skill name exposed to Claude is image-metadata-cleaner, matching the AIwork4me/image-metadata-cleaner repository and installation directory for official skill validation.

Use in Claude Code

Ask for privacy metadata cleanup of images you own or are authorized to process:

Clean privacy metadata from this folder of product images.

Or invoke directly:

/image-metadata-cleaner /path/to/folder --manifest

CLI examples

Preview planned outputs:

uv run --with pillow==12.2.0 python scripts/strip.py /path/to/images --dry-run

Clean one file and write photo-clean.png or photo-clean.jpg beside it:

uv run --with pillow==12.2.0 python scripts/strip.py /path/to/photo.png --manifest

Clean a folder into /path/to/images/metadata-cleaned/:

uv run --with pillow==12.2.0 python scripts/strip.py /path/to/images --manifest

Choose JPEG output explicitly:

uv run --with pillow==12.2.0 python scripts/strip.py /path/to/images --format jpg

Supported inputs

.png, .jpg, .jpeg, .webp, .bmp, .tiff, .tif

Verification scope

The script verifies that common metadata keys and common C2PA/JUMBF marker strings are not visible in the output file. This is a practical hygiene check, not a cryptographic guarantee.

It does not remove pixel-level watermarks, image fingerprints, external platform records, or any provenance signal outside the image file itself.

Development

Run tests:

uv run --with pillow==12.2.0 python -m unittest discover -s tests

License

MIT-0