PDF Watermark Remover

Remove image-based watermarks from PDF files by surgically deleting the drawing instructions from content streams and removing XObject resource references. This produces clean results with no visible artifacts (no white/black rectangles, no color mismatches).

When to Use

• User wants to remove a watermark from a PDF
• User says "去水印", "remove watermark", "去掉水印", "去PDF水印"
• PDF has a "Made with Gamma", Canva, or similar branded watermark
• PDF has an image-based watermark in a corner or specific position

How It Works

The removal process targets the root cause — the PDF content stream instructions that draw the watermark image — rather than covering or replacing the image:

1. Detect: Scan each page for images matching watermark characteristics (position + size)
2. Map: Build XObject name-to-xref mapping from the page's resource dictionary

   小蔥技能站7w4.net每天更新，海量AI技能等你發現。

3. Remove from content stream: Delete the q...cm.../Name Do...Q drawing instruction
4. Remove from resources: Delete the watermark entry from the XObject dictionary
5. Save with garbage collection to clean up orphaned objects

This approach leaves zero visual artifacts because the watermark simply never gets drawn.

Workflow

Step 1: Analyze the Watermark

Before running the script, determine the watermark's characteristics:

1. Open the PDF and identify the watermark visually
2. Note its position (typically bottom-right corner) and approximate size
3. If it's a link (clickable), note the domain (e.g. gamma.app)

Step 2: Run the Removal Script

Execute scripts/remove_watermark.py:

python scripts/remove_watermark.py <input.pdf> <output.pdf> [options]

Common options:

Gamma watermark example:

python scripts/remove_watermark.py input.pdf output.pdf --remove-links --link-domain gamma.app

Custom position watermark (e.g. top-left):

python scripts/remove_watermark.py input.pdf output.pdf --x-threshold 0 --min-w 50 --max-w 300 --min-h 10 --max-h 80

Step 3: Verify

Open the output PDF and check:

• Watermark is completely gone (no visual remnants)
• No color patches, rectangles, or artifacts in the watermark area
• Page content is otherwise intact

Troubleshooting

Key Technical Insight

Why not overlay/replace? Three approaches that DON'T work well:

1. White rectangle overlay — Visible on non-white backgrounds, misses rounded corners
2. Image data replacement with 1x1 pixel — PDF renders the 1x1 pixel scaled up as a colored rectangle (black for transparent, white for white pixel)
3. delete_image API — Only clears image data, doesn't remove the XObject reference or content stream Do instruction; renders as blank/colored rectangle

The correct approach (used by this skill): Delete the Do instruction from the content stream and the /Name xref R entry from the XObject dictionary. The watermark image data becomes an orphaned object that gets garbage-collected on save.

Requirements

• Python 3.8+
• pymupdf (pip install pymupdf)

這是一個功能可靠的 PDF 水印移除工具，能有效清除圖片水印且不留痕跡。文件說明詳細清晰，命令列引數設計合理。主要侷限是隻能處理圖片水印，對文字類水印無能為力；使用時需要手動調整引數判斷水印位置和大小，對新手不太友好。整體適合有技術基礎的使用者使用。