Help Center

Everything you need to know about removing watermarks with Unwatermark.

How It Works

Unwatermark uses a two-stage AI pipeline to remove watermarks from your files.

AI Detection
Local Removal
  1. AI Analysis -- Claude's vision model examines your image to find the watermark, understand what's behind it, and recommend the best removal technique.
  2. Local Removal -- The selected technique runs on the server with no additional API calls. Your files are processed and returned immediately.

Step-by-Step Guide

Step 1
Upload your file

Drag and drop a file onto the upload zone, or click to browse. Supports PNG, JPG, BMP, TIFF, WebP, PDF, and PPTX files up to 50 MB.

Step 2
Annotate or describe

For images, draw a rectangle around the watermark and/or describe it in plain language. For PDFs and PPTX, describe the watermark in the text field. Click Analyze with AI to detect and highlight it.

Step 3
Download result

For images, compare before/after with a slider. For documents, download your clean file. Use Try Again to refine, or Start Over for a new file.

Removal Techniques

The AI automatically selects the best technique for each watermark based on the background.

Solid Fill

Samples the color around the watermark and fills it in. Perfect for watermarks on solid-color backgrounds.

Gradient Fill

Interpolates colors from all four edges of the watermark region. Works well on gradient backgrounds.

Clone Stamp

Copies pixels from an adjacent area, mirrors them, and blends with a gradient mask. Good for simple textures and patterns.

Inpaint (LaMa)

Uses a state-of-the-art neural inpainting model to reconstruct content behind the watermark. Best for complex backgrounds with text, photos, or diagrams.

CLI Usage

Unwatermark also works as a command-line tool for batch processing.

$ unwatermark image.png $ unwatermark presentation.pptx -o clean.pptx $ unwatermark document.pdf --annotate "watermark in bottom-right" $ unwatermark photo.jpg --strategy solid_fill --no-ai

Run unwatermark --help for all available options.

Tips for Best Results

  • Draw a tight rectangle -- For images, the closer your selection fits the actual watermark, the cleaner the removal. Don't include too much surrounding content.
  • Describe both what and where -- Use both fields: "What does it look like?" (e.g., "gray NotebookLM text with icon") and "Where is it?" (e.g., "bottom-right corner, white on dark backgrounds"). The more detail you give, the better the AI performs.
  • Mention color variations -- If the watermark changes color depending on the background (e.g., "white on dark slides, gray on light slides"), mention that. The AI uses this to find it on every page.
  • Use high-res source files -- Higher resolution images give all techniques more surrounding context to work with.

FAQ

Is my file sent to the cloud?

Your image is sent to Claude's API for analysis only (to detect the watermark). The actual removal happens entirely on the server -- no image data is stored after processing.

What about PDFs and PPTX files?

PDFs are rendered to high-resolution images, cleaned, and reassembled. PPTX files have their embedded images processed in-place -- all slide layout, text, and formatting is preserved.

How does the AI decide which technique to use?

Claude's vision model examines the background behind the watermark. Solid backgrounds get a color fill, gradients get interpolated fills, and complex backgrounds (photos, text, diagrams) use LaMa neural inpainting for the best results.