Text, logos and which letters work
PDF has fonts every viewer carries: Helvetica, Times and Courier. Text set in them is stored as real text, which stays sharp at any zoom and costs a few hundred bytes a page. They cover the letters of Western European languages, digits and common signs, but not Greek, Cyrillic, Arabic, Hebrew or Asian scripts. For those, the tool draws the text on a canvas with your device's fonts, at three times the size it will have on the page, and adds it as a PNG with a see-through background. The page tells you which of the two is in use, and says so if your device has no font for the letters you typed.
Arabic and Hebrew are laid out by the browser, so letters join and run from right to left as they should. Text drawn as a picture cannot be selected. Real text can, so the word CONFIDENTIAL can show up when someone copies the page's text.
PNG and JPG logos keep their quality when they are ordinary and not larger than 2000 pixels on the long side: a JPG goes in with its own compressed data, so it is not compressed a second time, and the pixels of a PNG are stored without loss. Camera details and location that a JPG photo carries are cut out first, so they do not travel into every file you mark. Anything else (WebP, a huge photo, a JPG with a rotation flag) is redrawn at a size that is plenty for a mark on a page. A logo is stored once in the PDF, however many pages and copies use it.
Above the page or behind it
A page is drawn by running its instructions in order. A mark placed above is added after them, so it always shows. A mark placed behind is added before them, so the page's own drawing covers it. That keeps text easy to read, and it is why a behind mark works only where the page is see-through. Pages from a word processor usually have no background, so the mark shows between the letters. A page with a white or coloured rectangle over the whole sheet, a slide with a coloured background, or any scan, paints over it completely.
ClipLime finds out which pages do that by drawing each one small on a transparent canvas and counting how much is covered. The preview shows the result for the page you are looking at. With Show it on pages that cover it switched on, which it is by default, those pages get the mark above their content instead, and the result says how many were treated that way. If the check after saving cannot see the mark on a page it looked at, it says so and gives the likely reason, instead of reporting a success.
Placement, rotation and repeating
The mark is placed on the page as a reader sees it. A page stored sideways with a rotation entry, or cut to a crop box, still gets the mark in the corner you picked and upright. A file with pages of different sizes gets the same look on each, because the size is a share of the page width and not a fixed number of points.
Rotation turns the mark counter clockwise around its centre, from -180 to 180 degrees. Opacity runs from 5 to 100 percent: at 100 the mark covers what is under it.
Repeat across the page fills the page with copies in staggered rows that follow the rotation. The spacing is a share of the mark's own size, and the position control is not used then. A page can hold up to 400 copies. If your size and spacing would need more, the tool asks you to make the mark bigger or the gaps wider.
What a watermark can and cannot do
A visible watermark tells readers what a file is: a draft, a copy, confidential. It is not protection. It sits in the page like any other drawing, and a person with a PDF editor can select and delete it, or cover it. This tool does not encrypt the file, does not stop copying or editing, and does not turn the pages into pictures. If you need a mark that is harder to remove, the usual route is to flatten the pages into pictures, and this tool does not do that.
Everything else is copied unchanged: pages, fonts, forms, links, bookmarks and the document information. The PDF library writes the whole file again, so it is saved as PDF 1.7, the "fast web view" arrangement is lost, and a digital signature stops being valid. The mark is marked as an artifact of the page, so screen readers skip it. One mark adds roughly 250 bytes a page, and a tiled page with 50 copies about 800 bytes, plus the logo once.