image-provenance (863401402)

Properties

category: watermark-remover
maturity: emerging
approach: Zero-build, single-HTML-file, fully client-side (browser/WASM) toolkit combining an embedded C2PA WebAssembly validator and SynthID/frequency-domain analysis (65 features, FFT visualization) for detection, with a separate 'watermark disruption v2' module offering 8 techniques including genuine 2D-FFT phase perturbation at four preset intensity levels (light/recommended/strong/extreme), plus byte-level C2PA-metadata removal; images never leave the device.
access: open-source
owns: A dual-purpose, install-free browser tool that both analyzes an image's AI/provenance signals (C2PA, SynthID, EXIF/XMP/IPTC, frequency-domain artifacts) and offers built-in techniques to disrupt or strip those same signals, entirely client-side.
distinctFrom: Unlike noai-watermark, remove-ai-watermarks, or wmr (CLI/desktop tools using diffusion regeneration, inpainting, or alpha-blend inversion that require local install), this ships as a single zero-build HTML file running entirely in-browser with no upload and no local backend; unlike c2pa-viewer or origin-lens (read-only verifiers), it also actively disrupts the watermarks and metadata it detects via FFT phase perturbation rather than only reporting on them.
source: 863401402/image-provenance (GitHub)
notes: Quote translates to: '8 techniques (including genuine 2D-FFT phase perturbation) + 4 presets (light / recommended / strong / extreme)', from the README's 'watermark disruption v2' section. Repo also self-describes as an 'AI image detector and watermark disruption toolkit'. MIT license, 205 stars, created 2026-05-12, code last pushed 2026-08-11 (just before this lens's 2026-08-19 window; included as an untracked gap-fill rather than a confirmed new-in-window launch since it was not previously tracked in this catalogue). Repo description self-labels the project 'WIP'. Star count and feature list confirmed directly from the GitHub API and README; no independent accuracy figures are claimed or available to verify.

Provenance

Source: https://github.com/863401402/image-provenance

Properties

category

watermark-remover

maturity

emerging

approach

Zero-build, single-HTML-file, fully client-side (browser/WASM) toolkit combining an embedded C2PA WebAssembly validator and SynthID/frequency-domain analysis (65 features, FFT visualization) for detection, with a separate 'watermark disruption v2' module offering 8 techniques including genuine 2D-FFT phase perturbation at four preset intensity levels (light/recommended/strong/extreme), plus byte-level C2PA-metadata removal; images never leave the device.

access

open-source

owns

A dual-purpose, install-free browser tool that both analyzes an image's AI/provenance signals (C2PA, SynthID, EXIF/XMP/IPTC, frequency-domain artifacts) and offers built-in techniques to disrupt or strip those same signals, entirely client-side.

distinctFrom

Unlike noai-watermark, remove-ai-watermarks, or wmr (CLI/desktop tools using diffusion regeneration, inpainting, or alpha-blend inversion that require local install), this ships as a single zero-build HTML file running entirely in-browser with no upload and no local backend; unlike c2pa-viewer or origin-lens (read-only verifiers), it also actively disrupts the watermarks and metadata it detects via FFT phase perturbation rather than only reporting on them.

source

863401402/image-provenance (GitHub)

notes

Quote translates to: '8 techniques (including genuine 2D-FFT phase perturbation) + 4 presets (light / recommended / strong / extreme)', from the README's 'watermark disruption v2' section. Repo also self-describes as an 'AI image detector and watermark disruption toolkit'. MIT license, 205 stars, created 2026-05-12, code last pushed 2026-08-11 (just before this lens's 2026-08-19 window; included as an untracked gap-fill rather than a confirmed new-in-window launch since it was not previously tracked in this catalogue). Repo description self-labels the project 'WIP'. Star count and feature list confirmed directly from the GitHub API and README; no independent accuracy figures are claimed or available to verify.