The size is a percentage, and that is the point
The mark is sized as a share of each image's shorter side rather than in pixels. Set it once at 5% and it is the same visual weight on a 900px web export and on a 6000px original, which is what makes a batch look like it was done deliberately instead of one file at a time. A pixel size does the opposite: legible on the small file, a speck on the large one.
Everything else follows the same rule. The margin is a fraction, so the inset scales. The tile spacing is a fraction, so a tiled mark has the same density on every image in the batch.
Watermarks deter, they do not protect
A corner watermark is trivially cropped off. A semi-transparent one over a busy area is removable with content-aware fill in about thirty seconds, and machine-learning removal tools do it automatically. If a photograph must not be reused, do not publish it at full resolution; a watermark is a claim of authorship and a nuisance to the casual copier, not a lock.
Tiling raises the cost considerably, because there is nowhere to crop to and the removal has to work everywhere at once. The rows here are offset by half a step rather than sitting on a plain lattice, which makes an automated patch-and-repeat approach harder. It also makes the image less pleasant to look at, which is the trade: proof copies get tiling, portfolio pieces get a corner.
Common problems
- The watermark is invisible on some images. White text on a bright sky, or dark text on a shadow. There is no colour that works on every photograph; a mid-grey at higher opacity survives more backgrounds than white or black does.
- The font is not the one you designed with. The mark is drawn with the device's own sans-serif, because fetching a webfont to composite into a canvas would mean a network request, and this section promises none. For an exact typeface, render your mark as a PNG elsewhere and composite it in an editor.
- Nothing happened after adding files. Apply is the commit point. It runs the whole batch through the worker one file at a time, which is why the previews fill in progressively.
- The output lost its metadata. Watermarking re-encodes, so EXIF does not survive. Usually welcome: a watermarked photo you are about to publish should not be carrying GPS coordinates.
Frequently asked questions
Can I use an image as the watermark?
Not yet. Text covers the overwhelming majority of the use case: a name, a handle, a copyright line. An image mark brings its own scaling and alpha questions, worth solving properly instead of quickly.
Are my photos uploaded?
No. Every file is decoded, composited and encoded in a Web Worker on your device. That matters here specifically: uploading unwatermarked originals to a service in order to watermark them means the unprotected version has already left your machine.
What opacity should I use?
Between 40% and 70% for a corner mark: readable when looked at, ignorable when not. Tiled marks want to go lower, around 15% to 25%, because there are many of them and the total ink adds up fast.
Does it work on a phone?
Yes, though a batch of thirty full-resolution photos will use a lot of memory. Each file is processed one at a time and released afterwards for exactly that reason.