Watermark

A watermark is a marking embedded into a file to make the origin, authorship or usage status of a piece of content recognisable. The term comes from papermaking, where areas of differing thickness formed a motif that only became visible when you held the sheet up to the light. Digitally, the principle works in much the same way, except that the information sits inside the image, video, audio or PDF content itself. That is what sets it apart from metadata: a watermark survives an export or the resetting of file fields, because it is part of the content rather than merely a description of it.

You can broadly distinguish two families. A visible watermark places a logo, a note such as “sample” or a semi-transparent pattern over the motif. It has an immediate deterrent effect, but it interferes with the visual impact and is therefore better suited to previews, review cycles and sample files than to final publication. An invisible watermark alters pixel, frequency or audio values so slightly that you notice nothing; it can only be read out with the matching software. This approach relies on steganography, meaning the concealment of information inside other information. A particular variant is the forensic watermark: every recipient receives an individually marked copy, so that in the event of a leak you can trace which source the material came from.

What matters is how resilient the marking is designed to be. A robust watermark is meant to survive typical editing steps such as scaling, cropping, colour adjustment or heavy compression. A fragile watermark does precisely the opposite: it breaks at the first modification and thereby proves that a file has been tampered with. Both have limits. A screenshot, an aggressive crop or re-rendering a video can damage even robust markings, and a watermark does not prevent copying – it merely makes misuse traceable and less attractive. It therefore works best as a complement to solid documentation of your image rights and to Digital Rights Management measures, not as a substitute for them.

Since 2 August 2026, the transparency obligations under Article 50 of the EU AI Act have been binding. Providers of AI systems must mark their outputs in machine-readable form, for example through invisible watermarks or signed metadata. If you publish such content, however, that is not enough: deepfakes and realistic-looking AI images require a notice that is actually perceived – in other words a visible label in or next to the image. An invisible watermark alone does not satisfy this duty. At the same time, cryptographic provenance records are gaining importance, as we describe in our post on C2PA in DAM. Watermarks and provenance data complement each other here, because signature chains are quickly lost through screenshots or platform uploads.

Marking individual files by hand does not scale. In a Digital Asset Management system you store the marking as a rule instead and apply it on export through defined channels. You configure position, size, opacity and motif once, and the original file stays untouched. The same record can then deliver watermarked web previews for external partners and unmarked print data for your agency at the same time. Combined with rights and roles, you also control who is able to access the unmarked version at all – and it is exactly this interplay that makes watermarks practical in everyday work.

A watermark is not copy protection but an instrument for attribution and transparency. Visible markings help with previews, samples and mandatory labelling, while invisible and forensic variants support the tracing of leaks. What counts is deciding deliberately which assets are marked and in which form, because every marking costs visual impact and potentially reach. It is most effective where it is automated at the output channel, cleanly documented and embedded in an overall strategy of rights management, access control and provenance records.