863401402/image-provenance

Image Provenance is a browser based tool for checking whether an image was likely generated by AI, and for disrupting invisible AI watermarks hidden inside images. Everything runs entirely inside your browser, so images never leave your device. On the detection side, it checks for several kinds of AI origin signals: C2PA and Content Credentials, which are metadata standards used by companies like Adobe and some camera makers, known signatures left by specific AI image generators, and frequency domain analysis, a technique that looks at the underlying mathematical pattern of an image's pixels, which tends to differ subtly between AI generated and camera captured photos. Results come with a confidence rating, and only strong or medium signals are reported as an actual match. A separate metadata viewer displays all the hidden information stored inside an image file, including GPS location, shown with a privacy warning, a full editing history, and timestamps, drawing on the EXIF, XMP, IPTC, and ICC metadata formats. The conversion tools can strip out C2PA credentials, re encode an image through the browser's own canvas, inject realistic looking camera metadata from real camera models, and apply several watermark disruption techniques, including manipulating an image's frequency domain phase. The author states this disruption feature is meant for research into how well AI detection tools hold up, not for spreading disinformation or committing fraud. The whole project is built with plain HTML and JavaScript, with no framework, and relies on only two small external libraries, one for reading metadata and one for writing it. The README is written mainly in Chinese, with an English version also available, and the project is released under the MIT license.