Embed an invisible watermark in your image — and verify it later

The technique

What is LSB watermarking?

LSB stands for least-significant bit — the last bit of every byte. Each pixel in an image is stored as three color bytes (red, green and blue), and flipping the final bit of any of them changes that channel by at most 1 level out of 255. A shift that small is far below anything the human eye can register, which is exactly what makes it useful for hiding data.

This tool writes your text into those bits. The payload is prefixed with a short header — a format tag and the payload length — every bit is written five times for redundancy, and the bits are laid across the image in reading order: left to right, top to bottom, one bit per color channel. To read the mark back, the verifier walks the same path, takes a majority vote over each group of five copies, and rebuilds the original bytes.

The result is a hidden watermark that is invisible in normal viewing but readable by any software that knows where to look. Because it lives in the pixel data, it does not appear in an EXIF viewer or in a metadata cleaner's report — there is no metadata record to list.

Honest limits

What an invisible watermark survives — and what destroys it

An LSB watermark is only as durable as the exact pixel values that carry it. That makes its strengths and weaknesses unusually clear-cut.

It survives:

• Lossless re-saves. Saving the watermarked PNG again as PNG, BMP or TIFF preserves every bit exactly.
• Format conversion to PNG. Converting the marked file to any truly lossless format keeps the mark intact.
• Metadata edits. Adding, editing or stripping EXIF and XMP does not touch pixel data, so the watermark is unaffected.
• Cropping — sometimes. Bits are written starting from the top-left corner, so a crop that removes the top rows can cut off the format header and leave the rest unreadable. Short payloads sit near that corner; keep it intact and the mark survives.

It does not survive:

• JPEG or lossy WebP re-compression. Lossy codecs approximate pixel values, and the least-significant bits are the first detail they discard. One JPEG re-save erases the watermark completely.
• Resizing or resampling. Interpolation mixes neighboring pixels and scrambles the bit pattern.
• Filters, blurs, sharpening and added noise — anything that rewrites pixel values, including deliberate pixel-level perturbation such as a fingerprint reset.

If the verifier reports no watermark on a file you marked here, one of these almost certainly happened to the file in between.

A different layer

Invisible watermark vs. C2PA content credentials

An invisible watermark and C2PA content credentials solve related but different problems. C2PA is a signed provenance record attached to the file: it can state who created an image, with which tool, and whether AI generation was involved — and it is cryptographically signed, so tampering is detectable. But it lives in the metadata layer. Any metadata cleaner, including the one on this site, can strip it in one click, and many platforms remove it on upload.

An LSB watermark is the opposite trade. There is no signature, no standard and no tamper-proofing — anyone who knows the format can read or overwrite it — but because it sits in the pixels rather than in a metadata block, it does not disappear when metadata is cleaned. The two approaches complement each other: credentials give you tamper-evident provenance while the file travels intact, and a watermark gives you a quiet fallback mark of ownership. You can inspect the credential side of any image with the MetadataRemover C2PA credentials checker.