What a visible watermark achieves
A watermark on an image does not prevent theft, and it is worth being clear about that before deciding how heavily to apply one. Anyone determined to reuse an image can crop, clone out or reconstruct a watermark, and automated inpainting now does a competent job on marks placed over uniform backgrounds.
What a watermark reliably does is attribute. Images travel far from where they were published, stripped of context and links, and a mark carrying a name or domain survives that journey. Much unattributed reuse is casual rather than malicious — someone saving an image for a presentation or a post — and a visible mark converts a share into a credit.
It also signals that the image is claimed, which deters the low-effort end of the spectrum entirely. Someone assembling a page of stock imagery will skip a watermarked photograph for an unmarked alternative, which is often the whole objective.
Placement, opacity and the trade-off
Placement is the design decision that matters. A corner mark is unobtrusive and trivially cropped. A mark across the centre is nearly impossible to remove without damaging the image and equally hard to look past. A tiled or diagonal repeat across the whole frame sits in between and is the standard for proofs and preview copies.
The right choice follows from intent. For portfolio work meant to be enjoyed, a discreet corner mark at 30 to 50 percent opacity is appropriate — the image is the point. For client proofs before payment, an obtrusive repeated mark is appropriate, because the intent is that the proof is not usable as a final.
Contrast is what determines legibility, not size. White text at reduced opacity works over dark or busy areas and vanishes over a bright sky; a subtle drop shadow or a thin outline keeps a mark readable over both. Test on your lightest and darkest images rather than a mid-tone one.
Metadata and the invisible alternative
Embedded metadata is the complement to a visible mark. IPTC and XMP fields carry creator, copyright and contact information inside the file, and unlike a visible watermark they cost nothing visually. The catch is that they are easily stripped, deliberately or by a platform re-encoding on upload — which is why metadata is worth adding but not worth relying on.
For photographers the practical setup is both: metadata on everything, and a visible mark on anything published at usable resolution. Publishing at a resolution suited to screen display, rather than at full size, removes much of the incentive independently, since a 1,200-pixel image cannot be printed large.
If your goal is stamping a document rather than an image, PDF watermarking handles multi-page files directly and keeps the text selectable underneath.