Skip to content
Delete Image Metadata

Invisible Watermark

Hide a short message inside an image's pixels, and read it back later to identify your copy. Useful for tracking files you distribute — and fragile in ways worth understanding first.

  • .jpg
  • .jpeg
  • .png
  • .webp
Free
No account, no limit on files
0
Files uploaded — processing is local
4 formats
Lossless — pixels never re-encoded
Processed in your browser

How it works

  1. 1

    Choose Add or Check

    Add embeds a message; Check reads one back out.

  2. 2

    Load an image

    The pixels are decoded in your browser.

  3. 3

    Embed and verify

    The message is written, then read back before the download is offered.

How it works

Each pixel stores three colour values from 0 to 255. Changing the lowest bit of one of those values shifts it by one step — a difference of less than half a percent, invisible to the eye and invisible in any normal viewing.

This tool writes your message into those lowest bits: a length header, a marker so extraction can tell a real watermark from noise, then the UTF-8 text. A 1000 × 1000 image holds roughly 375,000 characters, so length is never the constraint in practice.

The alpha channel is deliberately left alone, because changing transparency can alter how an image composites over a background.

What it survives — and what destroys it

This is the part most invisible-watermark tools gloss over, so it is worth stating plainly. Least-significant-bit encoding is fragile by nature: anything that recomputes pixel values destroys it.

  • Survives: copying the file, moving it between devices, re-saving it as PNG, and any lossless transfer.
  • Destroyed by: JPEG re-compression, resizing, cropping, rotating, colour adjustment, and screenshotting.
  • Destroyed by: uploading to essentially any social platform, since they all re-encode images.
  • Not secret: anyone who knows the method can read the message, or erase it by randomising the low bits.

What it is genuinely useful for

The realistic use is marking copies of files you distribute directly, where you control the format. Send a client a PNG with their name embedded, and if that file resurfaces you can identify which copy it was. Mark review copies of a portfolio the same way.

It is not a rights-management system, it is not tamper-proof, and it will not survive the journey through a social network. Treat it as a serial number on a file, not as protection for an image.

Robust watermarking, and why this is not that

Watermarks designed to survive compression work in the frequency domain: they modify DCT or wavelet coefficients so the mark lives in the image's structure rather than in its least significant bits, and they trade a little visible quality for durability. Google's SynthID belongs to this broader family of pixel-level marking.

That is a materially harder problem and this tool does not implement it. Saying so is more useful than claiming robustness the code does not have — if you need a mark that survives a platform upload, this is not the right tool.

What this tool does

  • Every embed is verified by re-reading the marked pixels before you can download
  • Changes each colour value by at most 1 in 255, and never touches transparency
  • Handles any UTF-8 text, including non-Latin scripts and emoji

What it does not do

  • Destroyed by JPEG re-compression, resizing, cropping and platform uploads
  • Not a secret: anyone who knows the method can read or erase it
  • Only recognises watermarks written by this tool — it cannot detect other systems, including SynthID

Invisible Watermark

Hide a short message inside an image's pixels, and read it back later to identify your copy. Useful for tracking files you distribute — and fragile in ways worth understanding first.

Open the tool

Reviewed and updated . Behaviour described on this page is covered by the project’s automated tests.

Frequently asked questions

Will an invisible watermark survive Instagram or Facebook?

No. Those platforms re-encode every upload, and re-encoding recomputes pixel values, which destroys a least-significant-bit watermark. Nothing this tool produces will survive that.

Can I see the watermark in the image?

No. Each affected colour value changes by at most one step out of 255, which is far below what the eye can perceive, even on flat gradients.

Why is the output always a PNG?

Because PNG is lossless. Saving as JPEG would re-compress the pixels and immediately destroy the watermark you just embedded.

Can this detect SynthID or other AI watermarks?

No. It only recognises marks written by this tool. SynthID and similar systems use frequency-domain techniques and proprietary detectors, and no third-party tool can reliably read them.

Is this the same as steganography?

Technically yes — least-significant-bit steganography is the underlying method. The difference is intent: this is for marking your own files, with the message and the method both openly documented.

Can I use this to prove ownership?

It can help identify which copy of a file leaked, if the file was distributed losslessly. It is not legal proof of ownership on its own.

Does the message need to be secret?

It cannot be. Anyone who knows the method can read it, so treat the message as an identifier rather than a confidential note.

Why does the file get bigger?

Because the output is PNG. If your original was a JPEG, you are trading compression for the lossless format the watermark needs to survive.

Can I watermark the same image twice?

The second embed overwrites the first, since both use the same bits. Only the most recent message is recoverable.

Keep going

Related tools

Back to the main metadata remover