Invisible watermarking explained

How invisible watermarking works

The identifier is spread across the image rather than stored in one place, usually in the frequency domain rather than the raw pixels. Because it is distributed, cropping a corner does not remove it: the payload survives in whatever is left, up to a point.

That distribution is also why it survives a re-save. A visible mark occupies specific pixels, so deleting those pixels deletes the mark. An invisible one has no specific pixels to delete.

What an invisible watermark survives

The practical question is whether the payload still reads after the journey your files actually take, which is a different question from whether a visible mark could be erased. For a photographer whose work is scraped and reposted at full size, it holds. For an image that will be screenshotted on a phone and re-compressed twice, often it does not.

• Survives: re-saving, moderate compression, resizing, screenshots at full size, cropping up to roughly half the frame
• Degrades: heavy JPEG compression, small thumbnails, strong colour grading
• Destroys it: regenerating the image through an AI model, extreme crops, printing and re-photographing

Invisible watermarks are not C2PA or SynthID

Three different things get called invisible watermarking. C2PA is signed provenance metadata attached to a file, which platforms can read to label content as AI-generated. It travels with the file but is stripped when the file is re-encoded without the header.

SynthID and similar AI provenance systems embed a signal in the pixels of generated content, which is closer to a true invisible watermark and is what Veo and Gemini use. Neither is under your control as a creator: they identify the generator, not you.

What this guide means by an invisible watermark is the third thing: an identifier you embed in your own work so you can later prove it was yours.

When a visible watermark is still the right answer

An invisible mark deters nobody, because nobody can see it. If the goal is to make casual reuse feel wrong, a visible mark does that and an invisible one does not.

The two are not alternatives. The common setup is a visible signature for attribution on public posts, as covered in adding a watermark to photos, and an invisible payload underneath so a stolen file can be traced back if it matters later.

What to set up before you need an invisible watermark

Embedding is the easy half. The half people skip is keeping the unmarked master and a record of what was embedded in which file, because a payload you cannot read back proves nothing.

If provenance genuinely matters for your work, keep dated originals in a place you control. That is more useful in a dispute than any watermark, visible or not.