How to Watermark Photos Without Ruining Them

My first watermark was my name in the bottom-right corner, pure white, fully opaque, about a quarter of the frame wide. It looked awful. It also lasted about a week: someone cropped off the bottom 8% of the picture and reposted it without the name.

Both problems have the same root. I hadn’t decided what the watermark was for.

Attribution and deterrence are different jobs

If you want people who share your photo to be able to find you, the mark should be small, quiet, and in a corner. Nobody is trying to steal that shot. They just need a name to type into a search box.

If you’re sending proofs to a client, or posting portfolio work you don’t want lifted for someone’s Shopify banner, a corner mark is close to useless. Crop it, or run one pass of content-aware fill, and it’s gone in under a minute. What survives is a mark tiled across the whole frame, sitting on top of the subject’s face. That’s ugly on purpose. You don’t get to have both polite and effective.

Settings that hold up

Some numbers that beat “adjust to taste”:

• Corner signature: 15–25% of the image width, 35–50% opacity. Visible at thumbnail size, invisible to your eye after two seconds of looking at the photo.
• Tiled: 20–30% opacity, rotated around 30°, spacing 60–100%. Low contrast everywhere is tedious to paint out and easy to read past.
• White text over dark scenes, near-black over bright ones. Turn the shadow on either way. A white mark over a blown-out sky is a mark that isn’t there.
• Margin around 3% of the shorter side. Under 2% reads as a printing error, past 8% starts to look like a border you didn’t mean to add.

The batch problem nobody mentions

Here’s the part that changed how I think about this. In 2017, Tali Dekel, Michael Rubinstein, Ce Liu and William Freeman at Google Research published a paper on visible watermarks, and their point was that the weakness isn’t any single image. It’s the collection.

When the same mark appears in the same spot at the same opacity across hundreds of stock photos, you can estimate the watermark from the set itself and subtract it, automatically, without ever knowing what the original logo looked like. The consistency that makes batch processing convenient is the exact property that breaks it. Their suggested fix was to vary the mark slightly per image.

Which is uncomfortable advice for a tool that sells the ability to stamp fifty photos identically. So use the batch for speed, but if the whole set is going somewhere public and scrapeable, nudge the rotation and position between uploads. Ten seconds of work, and the assumption the attack depends on no longer holds.

Export choices that quietly cost you

A JPG source should go back out as JPG at around 90% quality. Saving a photograph as PNG instead makes the file four or five times heavier and looks the same.

If your source has transparency, don’t export to JPG at all. The format has no alpha channel, so every transparent pixel comes out white.

And whichever format you pick, the watermark is drawn into the pixels. There’s no layer to switch off later, so keep the originals somewhere you won’t overwrite them.

What a watermark is not

It’s a signature and a speed bump. If it ever comes to a takedown request, what carries weight is the untouched original, its EXIF and IPTC fields, and a date you can point at. The mark on the picture is what stops the casual reuse that makes up most of the problem.

The Image Watermark Tool does all of this in your browser — nine positions plus tiling, text or a PNG logo, and a zip for the whole batch, with nothing uploaded anywhere.