Gemini Watermark Tool: A Deterministic and Verifiable Approach to Image Watermark Removal

When working with images generated by Gemini models, one practical issue often appears quickly:
the semi-transparent watermark located in the bottom-right corner.

In slides, technical documents, UI mockups, diagrams, and screenshots, this watermark can interfere with readability and visual consistency. In many cases, users are not looking to “repair” or “guess” missing image content—they simply want the original pixels restored as accurately as possible.

Gemini Watermark Tool is a command-line utility designed specifically for this purpose.
Rather than relying on generative image repair or heuristic inpainting, it applies a mathematically deterministic reverse alpha blending algorithm to reconstruct the original pixels affected by the watermark.

This article explains what the tool does, how it works, how to use it safely, and where its limitations are—based entirely on the original project documentation.

What Problem Does This Tool Solve?

Gemini-generated images include a visible watermark, applied through alpha blending rather than hard pixel replacement.

This detail is important:

  • The watermark does not overwrite the original pixels
  • The underlying image information still exists
  • Pixel values are mathematically recoverable under the right conditions

Gemini Watermark Tool is built around this observation.
Its goal is not to invent missing content, but to reverse the blending operation and recover the original pixel values as precisely as possible.

Core Design Philosophy

Many image watermark removal approaches rely on techniques such as:

  • Generative inpainting
  • Content-aware fill
  • AI-based texture reconstruction

These methods often introduce side effects, especially in text-heavy or UI-based images:

  • Warped characters
  • Incorrect spacing
  • Altered line thickness
  • Invented visual details

This tool intentionally avoids those approaches.

No guessing. No hallucination. No generative reconstruction.

According to the project documentation, Gemini Watermark Tool is:

  • Deterministic
  • Non-generative
  • Based on mathematical inversion
  • Repeatable with identical input and output

The same input image will always produce the same result.

Key Features at a Glance

The tool focuses on practical, engineering-oriented functionality:

  • Batch processing of entire directories
  • In-place editing or explicit input/output paths
  • Automatic watermark size detection
  • Cross-platform binaries
  • Zero runtime dependencies
  • Single standalone executable

These features are designed for real-world workflows, not experimental image manipulation.

Supported Platforms

Pre-built binaries are provided for the following environments:

Platform Architecture
Windows x64
Linux x64
macOS Intel + Apple Silicon (Universal)
Android ARM64

All binaries are statically linked, requiring no additional runtime libraries.

Visual Examples

The original project includes multiple visual demonstrations comparing the original image, watermarked image, and restored output.

CLI Preview

CLI Preview

Processing Demo

Demo

Side-by-Side Comparison

Side by Side Comparison

The documentation emphasizes one particular use case:

Text-heavy slides and documents

In such cases, generative approaches often distort typography.
Reverse alpha blending preserves sharp edges, spacing, and character structure.

Getting Started

Simplest Usage: Drag & Drop (Windows)

For Windows users, the simplest workflow is:

  1. Download GeminiWatermarkTool-Windows-x64.exe
  2. Drag an image file onto the executable
  3. The watermark is removed in-place

No parameters or configuration required.

Command Line Usage

Minimal Mode (In-Place Editing)

GeminiWatermarkTool image.jpg

This command removes the watermark and overwrites the original file.

Warning: Always back up important images before using in-place editing.

Explicit Input and Output

GeminiWatermarkTool -i watermarked.jpg -o clean.jpg

This mode preserves the original file and is recommended for safer workflows.

Batch Processing (Directory Mode)

Entire directories can be processed at once:

GeminiWatermarkTool -i ./watermarked_images/ -o ./clean_images/

Supported image formats:

  • .jpg
  • .jpeg
  • .png
  • .webp
  • .bmp

How Watermark Size Is Detected

Gemini watermarks appear in different sizes depending on image dimensions.
The tool automatically determines which watermark size to remove.

Detection Rules

Image Dimensions Watermark Size Position
Width ≤ 1024 or Height ≤ 1024 48 × 48 Bottom-right, 32px margin
Width > 1024 and Height > 1024 96 × 96 Bottom-right, 64px margin

Examples

Image Size Detected Watermark
800 × 600 Small (48 × 48)
1024 × 1024 Small (48 × 48)
1920 × 1080 Large (96 × 96)

Manual override is available if needed.

Command Line Options

Option Short Description
--input <path> -i Input file or directory
--output <path> -o Output file or directory
--remove -r Remove watermark (default)
--force-small Force 48×48 watermark
--force-large Force 96×96 watermark
--verbose -v Detailed output
--quiet -q Errors only
--banner -b Show ASCII banner
--version -V Show version
--help -h Help message

How It Works: Reverse Alpha Blending

How Gemini Applies the Watermark

According to the documentation, Gemini applies its visible watermark using standard alpha blending:

watermarked = α × logo + (1 - α) × original

Here:

  • α represents transparency
  • logo is the watermark image
  • original is the original pixel value

Reconstructing the Original Pixels

By estimating or reconstructing the alpha map, the original pixel values can be recovered:

original = (watermarked - α × logo) / (1 - α)

This process:

  • Does not involve guessing
  • Does not use generative models
  • Produces deterministic output
  • Restores pixels mathematically

The README notes that the reconstructed alpha map is either correct or extremely close, enabling accurate restoration.

Frequently Asked Questions

Why does the image look almost unchanged after processing?

The watermark is semi-transparent.
If the underlying background is similar in color, differences may be subtle.

Recommendation:

  • Zoom to 100%
  • Inspect the bottom-right corner

The wrong watermark size was detected. What should I do?

Use a manual override:

GeminiWatermarkTool -i image.jpg -o output.jpg --force-small

or:

--force-large

I get a “file access denied” error.

Check the following:

  • The output directory is writable
  • The image file is not open in another application

Known Limitations

The tool is intentionally narrow in scope.
According to the documentation, it:

  • Removes only visible Gemini watermarks
  • Does not remove invisible or steganographic watermarks
  • Is designed for Gemini’s current watermark pattern
  • May not work if the watermark implementation changes

It is not intended as a general-purpose watermark removal solution.

Legal and Usage Notice

The project includes clear usage disclaimers:

  • Intended for personal and educational use
  • Users are responsible for legal compliance
  • The author does not encourage misuse
  • Provided “as is”, without warranty

Any legal implications depend on jurisdiction and usage context.

Building from Source (Advanced Users)

For those who prefer building manually, the project provides full instructions.

Required Tools

Tool Requirement
CMake 3.21+
C++ Compiler C++17
vcpkg Latest
Ninja Recommended

Supported platforms include Windows, Linux, macOS (Universal), and Android.

Project Structure

gemini-watermark-tool/
├── CMakeLists.txt
├── CMakePresets.json
├── vcpkg.json
├── src/
│   ├── main.cpp
│   ├── watermark_engine.cpp
│   ├── blend_modes.cpp
│   └── ascii_logo.hpp
└── resources/

Final Thoughts

Gemini Watermark Tool is not about removing a logo for visual convenience.
Its real value lies in its engineering approach:

  • Transparent methodology
  • Deterministic behavior
  • Mathematically grounded pixel recovery
  • Minimal assumptions

For document images, slides, UI assets, and diagrams—where accuracy matters more than aesthetic guessing—this approach offers a controlled and verifiable solution.