Watermarks Remover

Properties

category: watermark-remover
maturity: growing
approach: Agent skill plus a stdlib-only Python service that strips multi-vendor AI provenance marks from LLM-generated text (Claude/Anthropic, Gemini SynthID-Text, OpenAI, and open-weight-model watermarks, invisible-Unicode carriers, and statistical token-sampling marks) and from file metadata (C2PA/EXIF/XMP), with optional secondary backends for image (SynthID, StegaStamp, Tree-Ring, StableSignature) and audio (SilentCipher, AudioSeal, WavMark) watermarks.
access: open-source
owns: The first tracked watermark-remover whose primary, headline target is LLM-generated text watermarks (Claude, Gemini SynthID-Text, OpenAI, open-weight models), shipped as both an installable Python service and an agent skill for use inside coding-agent workflows; image and audio removal are optional add-on backends.
distinctFrom: The six already-tracked watermark-removers (noai-watermark, remove-ai-watermarks, reverse-synthid, wmr-gemini-watermark-remover, watermarkremover-ai, marknull) all target pixel- or frequency-domain image watermarks first. Watermarks Remover's core, headline capability is stripping text-generation watermarks (Anthropic's own scheme, Google's SynthID-Text) from raw LLM output, a modality none of the tracked removers lead with.
source: guillaumemeyer/watermarks-remover (GitHub)
notes: Quote is the repo's own README tagline with markdown bold markers removed and the em dash before 'for privacy' rendered as an ellipsis, per this run's no-em-dash copy rule; the source's own text uses an em dash at that point, not additional truncation of meaning. Created 2026-08-11; Show HN post 2026-08-28 (within this lens's window). 20,547 stars / 2,374 forks as of 2026-09-04 per the GitHub API, unusually rapid growth for a roughly three-week-old repo; MIT licensed, actively maintained (pushed same day as this check). A much smaller, overlapping-mechanism project (yasir-mo/AI-watermark-remover-GUI, 75 stars, MIT, created 2026-08-16, tested only against Claude text watermarking) was found and rejected as redundant with this larger, more comprehensive entry rather than added separately. The README and a vendor-notes reference claim Google retired its SynthID-text detector API access in August 2026; checked directly against Google DeepMind's own SynthID page, which shows no such retirement and still presents the SynthID Detector as active (early-tester waitlist), so that specific claim is omitted here and was not filed as an update against the tracked synth-id entry per the no-policy (source doesn't support the claim).

Provenance

Source: https://github.com/guillaumemeyer/watermarks-remover

Properties

category

watermark-remover

maturity

growing

approach

Agent skill plus a stdlib-only Python service that strips multi-vendor AI provenance marks from LLM-generated text (Claude/Anthropic, Gemini SynthID-Text, OpenAI, and open-weight-model watermarks, invisible-Unicode carriers, and statistical token-sampling marks) and from file metadata (C2PA/EXIF/XMP), with optional secondary backends for image (SynthID, StegaStamp, Tree-Ring, StableSignature) and audio (SilentCipher, AudioSeal, WavMark) watermarks.

access

open-source

owns

The first tracked watermark-remover whose primary, headline target is LLM-generated text watermarks (Claude, Gemini SynthID-Text, OpenAI, open-weight models), shipped as both an installable Python service and an agent skill for use inside coding-agent workflows; image and audio removal are optional add-on backends.

distinctFrom

The six already-tracked watermark-removers (noai-watermark, remove-ai-watermarks, reverse-synthid, wmr-gemini-watermark-remover, watermarkremover-ai, marknull) all target pixel- or frequency-domain image watermarks first. Watermarks Remover's core, headline capability is stripping text-generation watermarks (Anthropic's own scheme, Google's SynthID-Text) from raw LLM output, a modality none of the tracked removers lead with.

source

guillaumemeyer/watermarks-remover (GitHub)

notes

Quote is the repo's own README tagline with markdown bold markers removed and the em dash before 'for privacy' rendered as an ellipsis, per this run's no-em-dash copy rule; the source's own text uses an em dash at that point, not additional truncation of meaning. Created 2026-08-11; Show HN post 2026-08-28 (within this lens's window). 20,547 stars / 2,374 forks as of 2026-09-04 per the GitHub API, unusually rapid growth for a roughly three-week-old repo; MIT licensed, actively maintained (pushed same day as this check). A much smaller, overlapping-mechanism project (yasir-mo/AI-watermark-remover-GUI, 75 stars, MIT, created 2026-08-16, tested only against Claude text watermarking) was found and rejected as redundant with this larger, more comprehensive entry rather than added separately. The README and a vendor-notes reference claim Google retired its SynthID-text detector API access in August 2026; checked directly against Google DeepMind's own SynthID page, which shows no such retirement and still presents the SynthID Detector as active (early-tester waitlist), so that specific claim is omitted here and was not filed as an update against the tracked synth-id entry per the no-policy (source doesn't support the claim).