Adding watermarks to PDF documents involves direct manipulation of PDF content streams — the sequences of graphic operators that define everything visible on a page. Each PDF page has one or more content streams containing operators for drawing text, paths, images, and other visual elements. A watermark is implemented by injecting additional graphic operators into these streams, either before the existing content (underlay) or after it (overlay), to render the watermark text or graphic at the desired position.

The distinction between overlay and underlay watermarks is significant. An underlay watermark is rendered first, beneath all existing page content. This means the watermark text appears behind document text and images, which can make it less visible but ensures it never obscures important content. The implementation prepends watermark operators to the beginning of the page's content stream. An overlay watermark is rendered last, on top of all existing content. This guarantees the watermark is always visible but may partially obscure underlying text or graphics. The implementation appends watermark operators to the end of the content stream.

The PDF transparency model, defined in the PDF specification's section on transparency and blending, enables semi-transparent watermarks that are visible without completely blocking underlying content. Transparency is controlled through the graphics state parameter dictionary using the ca (non-stroking alpha) and CA (stroking alpha) parameters, with values from 0 (fully transparent) to 1 (fully opaque). A typical watermark uses an alpha value between 0.1 and 0.3 to create a subtle, non-intrusive appearance. The Extended Graphics State (ExtGState) resource must be defined in the page's resource dictionary and activated with the gs operator before rendering the watermark.

Positioning and rotating watermark text requires understanding the PDF coordinate system and transformation matrices. The text matrix (set with Tm operator) controls the position, scale, rotation, and skew of text rendering. A diagonal watermark rotated 45 degrees across the page center involves calculating the rotation angle, determining the page center from the MediaBox or CropBox, and constructing a transformation matrix that combines translation to the center point with rotation. The matrix components are computed as: a = cos(angle) * fontSize, b = sin(angle) * fontSize, c = -sin(angle) * fontSize, d = cos(angle) * fontSize, e = centerX, f = centerY.

Font handling for watermark text requires embedding or referencing a font resource. The watermark implementation must add a font dictionary to the page's resource dictionary, specifying the font name, encoding, and either referencing a standard 14 font (like Helvetica, which is guaranteed to be available in all PDF readers) or embedding a font file. The text rendering mode operator (Tr) can be set to 0 (fill), 1 (stroke), or 2 (fill then stroke) to control whether the watermark text appears as solid, outlined, or outlined-and-filled characters. Color is set using operators like rg (RGB fill color) or k (CMYK fill color) before the text operators.

A well-implemented watermark tool saves and restores the graphics state (using q and Q operators) around the watermark rendering to ensure that the watermark's transparency, color, and font settings do not affect the rendering of subsequent page content.