Understanding PDF watermarks
A stamp on every page.
What a PDF watermark actually does to the file, why it isn't security, and the small choices (opacity, rotation, layer) that make it look professional or homemade.
What the operation does.
A watermark adds new content to every page of an existing PDF — text or an image, usually centred or diagonal, usually semi-transparent. Under the hood it inserts a content stream alongside the page's existing one, drawn at a specified position with a specified opacity. The original page content is untouched; the watermark sits on top (or beneath) it as a separate drawing operation.
It isn't security.
A watermark on a PDF says "this document is a draft / confidential / for John Doe only" — it deters casual sharing, identifies leaked copies, signals intent. It does not prevent removal. Anyone with Acrobat Pro, Foxit, or a half-decent PDF library can strip a watermark in thirty seconds. For real protection use document encryption, access-controlled portals, or DRM systems. A watermark is a social control, not a technical one.
Three placement choices.
Beneath the page content (the watermark prints faintly behind the text — the "background" style). On top of the page content (overlays the document; harder to ignore but obstructs reading). As a header / footer band (specific edge strip, doesn't touch the main text area). Most leakage-tracking watermarks pick top-overlay at low opacity; most "draft" markers pick background. The right choice depends on whether you'd rather the watermark be unmissable or unobstructive.
A worked example.
A 50-page contract gets a watermark "DRAFT — 2026-05-13" rotated 45° at 20 % opacity, centred. The processor opens each page, finds its content stream, appends a new drawing instruction that pushes the current graphics state, sets opacity, translates and rotates the coordinate system to the page's centre, draws the text in a chosen font, then pops the state. The original content stream is untouched; reverting the watermark is just deleting the appended instructions.
Rotation, opacity, placement
45° rotation, 20 % opacity, page-centre origin
Three transform parameters fully describe a diagonal text watermark.
translate(W/2, H/2) → rotate(45°) → draw text
= Diagonal centred watermark
Fonts and Unicode.
A watermark embeds its font subset into the PDF. A Latin-letter watermark like "CONFIDENTIAL" needs only a few glyphs, adds maybe 5 KB. A CJK watermark needs the relevant code points and adds more. Many free PDF libraries default to the standard 14 base fonts (Helvetica, Times, Courier, Symbol, ZapfDingbats) which cover ASCII but nothing else. If your watermark contains accented characters or non-Latin scripts, the library needs explicit font embedding.
Image watermarks.
A company logo or a "VOID" stamp works as a PNG with transparency. The processor embeds it as a PDF image object once and references it on every page — repeating the image bytes per page would bloat the file. The same per-page placement logic (translate, scale, rotate, opacity) applies. PNG with a working alpha channel is the right format; JPEG is wrong because it has no transparency and the rectangular background would show up.
What runs in the browser.
PDF watermarking in the browser uses pdf-lib or pdfjs-dist to load the file, manipulate the page content streams, and re-serialise. The performance scales with page count and font subset size — a 200-page document with an embedded CJK font is several seconds; a 5-page contract with ASCII text is near-instant. No upload required; the file never leaves the tab.