The picture stays on your machine
Decode, draw, write text, encode: four operations, all of them in the tab. Nothing is uploaded, which matters here more than it does for most tools because the files people watermark are often the ones they are least comfortable handing to a stranger — identity documents, contracts, unreleased work. Watch the network tab, or disconnect entirely and watermark a document offline.
What a watermark is for
It is not protection. A mark in a corner is cropped off in seconds; a translucent mark across the frame can be attacked with inpainting tools that are now very good. Anyone determined will get a clean copy.
What it does achieve is worth having anyway. Provenance travels with the image, so a screenshot passed through five accounts still carries your name. Removal becomes a deliberate act, which changes the character of any dispute about it. And on a copy of a document, a purpose written across the face — who it went to and what for — makes the copy awkward to submit anywhere else, which is the entire point when a bank asks you to send one.
Placement conventions
| Purpose | Position | Size and opacity |
|---|---|---|
| Photography credit, portfolio | Bottom right | 3-5%, 35-50% |
| Publication or broadcast still | Top left or top right | 4-6%, 60-80% |
| Stock preview, proof copy | Tiled, or dead centre | 8-12%, 25-40% |
| Copy of an identity document | Tiled, or dead centre | 10%+, 60%+ |
| Draft or internal-only marking | Dead centre | 12%+, 20-30% |
Bottom right is the default because western reading order puts the eye there last, so it registers without competing with the subject. Tiling is the choice whenever cropping is the threat you care about, since there is no rectangle you can cut that does not contain a mark.
Choosing the numbers
Size is a percentage of image width rather than a pixel value, so the same settings look consistent across a 900 px screenshot and a 6000 px camera file. Three to six percent is a credit line; ten percent or more is a mark you are not meant to ignore.
Opacity between 30 and 50 percent stays readable without destroying the photograph underneath. For a document copy, push it to 60 or higher — legibility of the mark matters more than the appearance of the page. White text with the shadow on works over almost anything, because the shadow supplies an edge wherever the background matches the text. Black text is only better on consistently pale material, and if you turn the shadow off you should check the result against the lightest area in the frame.
Marking a copy of a document
Institutions that ask for a scan of an ID often ask you to write the purpose and date across it. Something like "For account opening at X Bank, 2026-08-28" tiled over the face does the job: the copy is still readable for the purpose it was sent for, and visibly useless for anything else.
The watermark is only half of it. Anything the recipient did not ask for should be covered rather than merely written over, and translucent text is not covering. Run the file through the pixelate tool first for the fields that should not be legible at all, then watermark the result. Doing it in that order means the mark sits on top and cannot be lifted to reveal what is underneath — because there is nothing underneath any more.
Questions people ask
Can I use a logo image instead of text?
Text only. An image watermark needs a second file loaded, scaled, positioned and composited, and doing that well needs a drag interface rather than a form. If you have a logo, the practical route is to place it in an editor once, save that as a template, and reuse it. For a text mark that reads as a brand, a distinctive short string in the corner does most of the same work.
Can the watermark be removed later?
Not from this file, and not by you. The text was composited into the pixel values and the original pixel values are gone from the output. This is what makes a watermark worth anything, and it is why you keep the unmarked original in a place you will find it again. Retouching tools can approximate a removal, which is a different thing from getting the original back.
Why can I not choose the font?
The text is drawn with the system sans-serif in bold, whatever that resolves to on the machine doing the work. Loading a custom font into a canvas means fetching a font file, which would break the guarantee that nothing leaves your browser and nothing is fetched into it. If typography matters to the mark, an editor is the right tool.
The tiled version is extremely slow on a big photo.
Tile count scales with area divided by text size, so a 6000 px photo with a small mark can end up drawing many thousands of strings. Raise the size percentage — which both reduces the count and produces a more useful mark — or resize the image before watermarking if the output does not need full resolution.