watermarks-remover

A closer look at watermarks-remover

watermarks-remover is a privacy-first open-source app that strips AI watermarks from content the user owns, at version 0.7.0 on the repository page read. It covers file metadata such as C2PA, EXIF, XMP, and document properties across images, documents, audio, and video, plus statistical text watermarks and vendor provenance surfaces from Claude, Gemini SynthID-Text, and OpenAI. It runs as an agent skill over a local HTTP service, as a Docker core service, or through CLI scripts such as clean_file.py, with hook modes that check findings or clean files in place.

Editorial guidance based on published documentation. The workflow below is a suggested starting point, not a measured product result.

When to choose watermarks-remover

• Developers using Claude Code or Codex who want provenance marks cleaned from their own files, since the project ships agent skills, a local service, and a hook with check and clean modes.
• Teams that want consistent gates, since detection is shared across the hook, the pre-commit gate, and the CI SARIF export.

When to choose something else

• Users who want to strip marks from content they do not own, since the project positions itself as stripping AI watermarks from content you own.
• Users who need a standalone desktop app with a graphical interface, since the documented paths are an agent skill, a local HTTP service, and CLI scripts.

A practical starting workflow

1. Install the agent skill for Claude Code or Codex, or start the local service, including the Docker core service on 127.0.0.1:8765.
2. Point the tool at a file the user owns, in check mode to report findings or clean mode to strip marks.
3. Run clean_file.py on the file to produce a cleaned output file, with clean mode swapping in the result only on a real difference.
4. Wire the same detection into a pre-commit gate or CI SARIF export so hooks, gates, and CI agree on what counts as actionable.

Check before committing

• The license covers the software only, so hosting, model API use, and local compute are separate costs, not free.
• Clean mode writes to a sibling temp file and swaps only on a real difference, so already-clean files keep their modification time and do not retrigger file watchers.
• Detection is shared across the hook, the pre-commit gate, and the CI SARIF export, so all three agree on what counts as actionable.

Common questions

Compare approaches for this task

• wasitai — Users who want a quick check on a single image, since wasitai offers drag-and-drop upload with a detailed report and confidence score.; Developers and teams who want detection in their own tools, since wasitai lists API and browser extension access.
• Claude — Working through source material, developing structured drafts, and iterating on code or documents.