It deters, it does not protect
Worth being clear about this before anything else, because the expectation is usually wrong.
A watermark raises the effort required to reuse an image and marks where it came from. It does not prevent copying. Anyone determined can remove one, and the tools for doing so have improved considerably — content-aware fill handles a corner mark in seconds, and modern inpainting models handle a good deal more than that.
So the useful question is not "will this stop theft" but "does this make casual reuse less likely and attribution more likely". The [msm_tool slug="watermark-generator"] is good at both of those and cannot do the first.
Corner or tiled — the decision that matters
A corner mark is removed with a crop. One rectangular selection and it is gone, along with a small amount of the image nobody will miss. It is appropriate for attribution on images you are happy to have shared: a photo credit, a site name, a handle.
A tiled diagonal pattern cannot be cropped away, because it covers the whole surface. Removing it means damaging the image it sits on. This is what proof and preview images use, and it is the only version that meaningfully protects anything.
If the image is a paid asset shown as a preview, tile it. If it is a portfolio piece you want people to share, a corner mark is more considerate and does the attribution job.
Opacity: twenty to forty percent
Below twenty the mark disappears on busy images — and busy images are precisely the ones people want to take. Above forty the image stops being worth looking at, which defeats the purpose of showing it.
Judge it on the actual photograph rather than on a flat background. A mark that reads clearly on a preview panel can vanish entirely over a detailed area, and the whole point is that it survives the parts of the image that matter.
Size and scale
The font size scales with the image, so a watermark that looks right on a preview looks right on the full-resolution file. That matters more than it sounds: a fixed pixel size tuned on a 1200-pixel preview is nearly invisible on a 6000-pixel original, and someone downloading the large version gets an unmarked image in practice.
What to put in it
Short. A name, a handle or a domain. Long text becomes unreadable at any opacity that keeps the image usable, and a paragraph of copyright notice is both ineffective and unpleasant.
A domain is usually the most useful, because it survives the image being reposted without context — someone who sees it knows where to look. A social handle works if the handle is stable; domains outlive platforms.
Format is preserved
A PNG comes back a PNG with its transparency intact, and a JPEG is not needlessly converted into something several times larger. That avoids the common annoyance of watermarking a logo and getting back a file with a black background.
Watermark the copy, always
The mark is drawn into the pixels and cannot be undone. Work from a duplicate and keep the original clean — you will want it when the watermark design changes, when a client asks for an unmarked version, or when the mark turns out to sit over the one part of the image that matters.
What actually protects images
If the goal is genuinely to prevent unauthorised use, watermarking is the weakest of the available measures. Stronger ones, in rough order:
- Publish at a lower resolution. A 900-pixel image is fine on screen and useless in print, and this stops commercial reuse more effectively than any mark.
- Register the copyright where your jurisdiction offers it. This is what gives a takedown or a claim actual force.
- Keep the metadata on published copies — author, copyright, contact. It is stripped easily but its presence establishes intent.
- Reverse image search periodically, so you know when it happens.
A watermark sits alongside these as a visible marker. On its own it is a deterrent, and deterrents work on the casual and not on the determined.