Image Watermark Remover

Clean images of embedded metadata and visible watermarks

An image watermark comes in two entirely different forms, and they need different treatment. A visible watermark is painted into the pixels - a logo across a stock photo, or the small badge some AI image generators burn into a corner. An invisible watermark is data riding alongside the pixels: metadata blocks such as EXIF, XMP, IPTC, and C2PA Content Credentials. This tool handles both, by separate mechanisms.

Content Credentials are provenance records defined by the C2PA specification, a standard developed by the Coalition for Content Provenance and Authenticity - an initiative backed by Adobe, Microsoft, the BBC, Sony, and Leica. A C2PA manifest cryptographically records who created an image, which tool made it, what edits were applied, and whether a generative model was involved. It travels invisibly inside the file, so a screenshot-free copy of your image can still announce the software, device, and editing history behind it.

Detection works by reading the raw bytes of the file you drop in. In a JPEG, a C2PA manifest lives in an APP11 application marker segment; in a PNG it is carried in a caBX ancillary chunk. The tool scans for those signatures and reports what it finds before changing anything.

How stripping and inpainting actually work

Metadata removal exploits a deliberate property of the HTML5 Canvas API: when an image is drawn to a <canvas> and re-exported withtoDataURL(), the browser writes out only the decoded pixel buffer. Every ancillary block - EXIF camera model and GPS coordinates, XMP editing history, IPTC caption fields, ICC profiles, and the C2PA manifest - is discarded because it was never part of the pixel data in the first place. The output is a clean PNG with nothing but image content.

For visible marks, drawing a rectangle over the area triggers a simpleinpainting pass: the algorithm samples colors from the pixels bordering your selection and blends them inward to fill the region. This is the classic approach, and its limits follow directly from the method - it reconstructs smooth gradients well and structure poorly. A logo over open sky, a studio backdrop, or a solid wall usually vanishes cleanly; the same logo over foliage, a face, or printed text will leave a visible smear, because the surrounding pixels contain no information about the pattern that should continue underneath.

One boundary worth stating plainly: removing a watermark does not remove the copyright. This tool is intended for images you own or have the right to edit - stripping provenance data from your own photos before publishing, or cleaning up output from a generator you licensed. It is not a way around someone else's licensing terms.

Detection, stripping, and inpainting all execute in your browser through the Canvas API. The file is read with FileReader and never uploaded, so confidential or client-owned images stay on your machine. Output is written as PNG to avoid a second round of lossy JPEG compression.