agentleFS
jqaisystemsjqaisystems/jqai-ai-skills/skills/remove-ai-image-metadata/SKILL.md

remove-ai-image-metadata

Skill3 starsChanged 6 months agoMITView on GitHub

Inspect, remove, and verify embedded AI-generator or provenance metadata in image files without changing encoded picture data when the format permits. Use for any image-metadata cleanup request involving GPT Image, OpenAI, DALL-E, C2PA, Content Credentials, generative-AI labels, EXIF/XMP generator fields, or similar provenance in individual images or folders, including PNG, JPEG/JPG, WebP, GIF, TIFF, HEIC/HEIF, AVIF, SVG, and other image formats.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 8 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Remove explicit embedded AI provenance or generator metadata using a format-aware workflow.

    line 8
  2. Distinctive:

    Prefer verified copies when replacement was not explicitly requested:

    line 23
  3. Very specific:

    Use --in-place only when the user wants the originals changed.

    line 29
  4. Very specific:

    Add --all-metadata only when the user asks to remove all removable non-rendering metadata.

    line 30
  5. Distinctive:

    Use the bundled script for byte-preserving PNG, JPEG/JPG, WebP, and GIF cleanup.

    line 38
  6. Distinctive:

    If no safe handler is available, stop and report the unsupported format.

    line 40
  7. Distinctive:

    Never claim a cleaned image is undetectable or guaranteed not to be classified as AI-generated.

    line 45
  8. Distinctive:

    Default JPEG mode refuses matched EXIF blocks that need selective tag editing, preserving orientation and co-located fields.

    line 48
  9. Distinctive:

    Keep backups outside the upload/publish folder so tagged originals are not selected accidentally.

    line 49
  10. Distinctive:

    Stop on malformed containers, failed integrity validation, changed encoded-image hashes, or incomplete post-clean inspection.

    line 50

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Remove AI Image Metadata

Remove explicit embedded AI provenance or generator metadata using a format-aware workflow. Preserve encoded image content and rendering data whenever possible.

Workflow

  1. Resolve the exact files or folder in scope and identify each real file format from its contents.
  2. Inspect before writing. For PNG, JPEG/JPG, WebP, and GIF, use the bundled cleaner without a write option:

    python "<skill-folder>\scripts\clean_image_metadata.py" "<file-or-folder>" --recursive
  3. Review every proposed removal. Default mode removes supported C2PA/Content Credentials blocks and metadata fields that contain known AI-generator markers. JPEG JUMBF requires a metadata-aware editor because a segment can contain unrelated or split records. Standard JPEG, WebP, and PNG iTXt XMP uses field-level edits, preserving unrelated sibling attributes and records. Photoshop support is limited to parsed Unicode text fields in descriptor-free version 6 slices; matching strings become same-length spaces. Unsupported matched structures and extended XMP stop for a metadata-aware editor. See selective handling.
  4. Choose a write mode:

    • Prefer verified copies when replacement was not explicitly requested:

      python "<skill-folder>\scripts\clean_image_metadata.py" "<file-or-folder>" --recursive --output-dir "<clean-folder>"
    • Use --in-place only when the user wants the originals changed. Add --backup-dir "<folder-outside-the-publish-folder>" when a backup is useful.
    • Add --all-metadata only when the user asks to remove all removable non-rendering metadata. Preserve color profiles and functional animation data.
  5. Confirm every changed file reports PASS. The cleaner reparses the written container and confirms that encoded PNG, JPEG, WebP, or GIF image payloads are unchanged.
  6. Inspect the cleaned output again when extra verification is appropriate.
  7. Report the number of files processed, exact output path, formats handled, and metadata removed.

Format Routing

  • Use the bundled script for byte-preserving PNG, JPEG/JPG, WebP, and GIF cleanup.
  • For TIFF, HEIC/HEIF, AVIF, SVG, PSD, or another format, inspect available format-aware tools such as ExifTool or c2patool. Work on copies unless in-place replacement was requested. Verify the result with the same tool plus a decoder.
  • If no safe handler is available, stop and report the unsupported format. Do not generically re-save or convert the image merely to strip metadata; that may alter pixels, animation, orientation, color, or compression.

Operational Rules

  • Process only images you own or are authorized to modify. Do not remove provenance metadata to misrepresent an image's origin or bypass a platform's disclosure requirements.
  • Never claim a cleaned image is undetectable or guaranteed not to be classified as AI-generated. State only which explicit metadata was removed. Pixel-based classifiers may still guess.
  • Preserve encoded picture data, ICC profiles, gamma/chromaticity information, JPEG color-transform markers, animation control, and other rendering-related data.
  • In default mode, preserve unrelated copyright, authorship, descriptions, and location metadata unless the user requests broader removal.
  • Default JPEG mode refuses matched EXIF blocks that need selective tag editing, preserving orientation and co-located fields. Use a metadata-aware editor instead. Explicit all-metadata mode is broader and can remove EXIF orientation; check display orientation before choosing it.
  • Keep backups outside the upload/publish folder so tagged originals are not selected accidentally.
  • Stop on malformed containers, failed integrity validation, changed encoded-image hashes, or incomplete post-clean inspection.

Common Invocations

Inspect supported images:

python "<skill-folder>\scripts\clean_image_metadata.py" "C:\path\images" --recursive

Clean originals after explicit confirmation:

python "<skill-folder>\scripts\clean_image_metadata.py" "C:\path\images" --recursive --in-place

Create clean copies:

python "<skill-folder>\scripts\clean_image_metadata.py" "C:\path\images" --recursive --output-dir "C:\path\clean-images"

The bundled cleaner uses only the Python standard library.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.