Guide
PDF Watermark overlays a piece of text (including Chinese, and multiple lines) or an image onto every page of a PDF, or onto a page range you choose, at an adjustable opacity and angle — all inside the browser, with no upload.
Updated 2026-09-093 sources10 min read
PDF Watermark overlays a piece of text (including Chinese, and multiple lines) or an image onto every page of a PDF, or onto a page range you choose, at an adjustable opacity and angle — all inside the browser, with no upload.
It solves the “mark the file before it goes out” problem: writing “Draft · not in force” on a contract draft, tiling “Confidential” across internal material, or putting the company logo on a sample. Unlike “PDF Page Numbers”, a watermark is a translucent layer drawn over the body text, sitting in the centre of the page or tiled across it rather than being a line of small type in the margin.
The boundary matters too: a watermark is an added image object. It hints and it traces, but it cannot prevent copying and it does not replace encryption. To restrict opening or printing, pair it with “PDF Protect”.
1-3,5,8-10, or click pages in the middle area and let “Selected pages” fill the range in.<original name>_watermark.pdf and downloads automatically; the result card offers “Download again” or “Add to workspace” to keep going, for example to add page numbers or compress.
The input was a six-page English contract with vector text only, contract-6p.pdf (10.1 KB), where page 4 carries /Rotate 90 (displayed landscape). Two runs were measured:
Settings Pages Output size Notes
Default: Confidential · 48 pt · 30% · −30° All 59.7 KB One PNG embedded for portrait and
· centred landscape pages, 2 in total
Two lines "Confidential / Internal review 1-3 76.5 KB Tiled, gap 120 pt; only 1 PNG
only" · 36 pt · 20% · −30° · tiled embeddedAlmost all of the growth comes from the embedded PNG (the text is drawn as an image — see the next section) and hardly depends on the page count: whether the watermark covers 6 pages or 3, pages sharing one rotation angle share one embedded image.
The page and object structure of a PDF is covered in the “How it works” section of “PDF Merge” (see here); this section only covers what is specific to watermarks — why the text becomes an image first, and how it is placed on the page.
Writing text directly into a PDF needs a font: none of the 14 standard fonts contains CJK glyphs, and embedding a CJK font means handling glyph subsets, encoding maps and licensing. This tool sidesteps the problem: the browser canvas draws the text in the font, size and colour you chose onto a transparent canvas, and the canvas is then encoded as a PNG and embedded into the PDF. The “kai” style you pick in the panel is the one on your computer, and the recipient sees exactly the same thing, because the PDF contains only an image.
The drawing resolution is 3 pixels per pt: 48 pt text is drawn 144 px tall, line spacing is 1.25× the font size, and a 20% padding is left around the edges. The 3× oversampling keeps edges smooth when a reader zooms to 200%–300%, and going higher would only add size for nothing.
The angle is not written into a PDF transformation matrix — it is baked straight into the image: the transparent canvas is rotated, and the output canvas is the bounding box that contains it (width = w·cosθ + h·sinθ). Everything after that only has to place a tidy rectangle.
A PDF page can carry /Rotate 90/180/270, meaning the whole page is rotated for display. Drawn directly, the watermark would rotate along with the page. The tool uses “slider angle − page rotation” as the effective baking angle and converts the visual coordinates back into page coordinates, so the landscape page 4 ends up looking the same way round as the portrait pages. Pages that share a rotation angle share one embedded image, so more pages only add one more copy of it.
30% opacity does not fade the PNG's pixels; it sets the constant alpha of the graphics state at draw time (the /ca entry in §8.4.5 of the PDF specification), and the reader blends the image with the body text underneath. It also means that selecting this image object in an editor and deleting it removes the whole watermark — a watermark is not tamper-proofing.
Centred: x = (page width − watermark width) / 2, y = (page height − watermark height) / 2, where the width and height are those of the bounding box after rotation.
Tiling starts half a watermark outside the top-left corner of the page, steps horizontally by watermark width + gap and vertically by watermark height + gap, and continues until it runs past the page. Starting outside is what covers all four edges instead of leaving a blank band around them. The default 48 pt “Confidential” has a bounding box of roughly 140 × 126 pt at −30°; with an 80 pt gap an A4 portrait page fits about 4 columns × 5 rows.
An image is first fitted into a canvas whose longest side is at most 2000 px (larger images are scaled down proportionally), then it goes through the same rotation baking and embedding. The displayed width is converted from “a percentage of the page width”, so the same logo occupies the same share of an A4 and an A3 page. A PNG with an alpha channel keeps its transparency; a JPEG is a solid rectangle.
2- (page 2 to the end). The page-range syntax matches “PDF Split” (see here).1-3,5,8- style ranges and “Selected pages” are supported; page numbers follow the current arrangement, including pages deleted or reordered in the workspace./Rotate are compensated automatically so the watermark keeps a consistent visual direction.When legal sends a contract that is still being amended, add a centred “Draft · subject to the signed version” at 30% opacity so it cannot be mistaken for a final copy; issue the final version as a separate file with no watermark.
Send the same training material to several departments, each tiled with “Confidential · Marketing only”. If the file leaks, the watermark says where it came from. Use Enter for multiple lines.
An annual report cover carries a full-bleed design and should not have text over it. Set the page range to 2- and tile “Internal” from page 2 onwards, leaving the cover untouched.
Yes. The text is drawn as an image using the fonts in your browser and then embedded, so as long as the preview shows the right characters the output does too, without depending on the recipient's fonts.
Only one PNG is embedded per distinct page rotation angle, and each page references it from a drawing instruction. The size depends mainly on the font size and the length of the text: bigger type and more lines make a bigger PNG.
No. A watermark in someone else's file may be an image, text or a vector object, and its relationship to the body varies; removing another party's watermark usually raises copyright and contract issues as well, so ask the author for an unwatermarked copy.
No. Adding anything to a signed PDF changes the file, and verification will report that the document has been modified. The correct order is to watermark first and sign afterwards.
The image has no alpha channel (a JPEG, for instance), so its white background is drawn as content. Save the logo as a PNG with a transparent background.
Text drawing, image handling and PDF reading and writing all happen inside your browser; the file never leaves your device and the tool works without a network connection. See the privacy policy for more.
/ca constant alpha), §8.9.5 Image Dictionaries (/SMask), §7.7.3.3 Page Objects (/Rotate): https://opensource.adobe.com/dcs/pdf/PDF32000_2008.pdf（访问日期：2026-09-09）CanvasRenderingContext2D.measureText() — measuring text width to size the watermark canvas: https://developer.mozilla.org/zh-CN/docs/Web/API/CanvasRenderingContext2D/measureText（访问日期：2026-09-09）PDFPage.drawImage(image, { x, y, width, height, opacity }): https://pdf-lib.js.org/docs/api/classes/pdfpage#drawimage（访问日期：2026-09-09）Updated 2026-09-09
Text (CJK supported) or image watermark with adjustable opacity, rotation and tiling
Everything runs locally in your browser; files are never uploaded to a server