allenk/GeminiWatermarkTool

VEO / Gemini 3.5 Nano Banana Pro watermark maintenance tool

Gemini Watermark Tool – What It Is

GeminiWatermarkTool is a small, cross‑platform program that strips the visible watermark Google adds to images (and, via a companion binary, to videos) generated by its Gemini AI models. It does this with a deterministic reverse‑alpha‑blending algorithm rather than generic generative inpainting, so the original pixel data is recovered as accurately as possible. An optional lightweight neural‑network denoiser (FDnCNN) can clean up any residual artifacts.

Key Features (as described in the README)

Category Details
Core algorithm Reverse‑alpha blending using calibrated 48×48 and 96×96 masks for the Gemini watermark.
Watermark profiles Supports the newer Gemini 3.5 layout (36×36 or 96×96 logos) and falls back automatically to the legacy pre‑3.5 layout.
Detection Three‑stage Normalized Cross‑Correlation (NCC) with confidence scoring; skips non‑watermarked images.
CLI Single‑file, batch‑directory, and multi‑file modes. Automatic legacy fallback, configurable --threshold, --force, --legacy, and distinct exit codes (0 = ok, 1 = skipped, 2 = error).
GUI Drag‑and‑drop desktop app with real‑time before/after preview, custom region selection, in‑painting (NS, TELEA, Soft), AI denoise, batch thumbnails, and keyboard shortcuts.
AI‑assisted cleanup FDnCNN (20‑layer convolutional network) run via NCNN with Vulkan GPU acceleration; CPU fallback if no GPU.
Video support Separate VeoWatermarkRemover binary (v0.6.1) that applies the same reverse‑alpha math frame‑by‑frame, with adaptive per‑frame alpha, audio passthrough, and optional AI denoise.
Cross‑platform Pre‑built portable executables for Windows, Linux, macOS (Intel + Apple Silicon) and Android CLI.
Agent integration MCP server and Claude Code skill (gwt‑integrations) expose four tool calls (remove_watermark, detect_watermark, batch_process, get_tool_info) for AI‑agent pipelines.
License MIT – free for commercial use, with attribution requirements for redistribution.

Typical Use‑Cases

  1. Cleaning up Gemini‑generated images before publishing or archiving, preserving the original detail without relying on generative fill.
  2. Batch processing large folders of screenshots or exported assets where some files may not contain a watermark.
  3. Integrating into automation scripts (e.g., CI pipelines, content‑moderation bots) using the CLI’s exit codes and the MCP‑exposed API.
  4. Video post‑processing for recordings that contain the Gemini 3.5 “diamond” watermark, using the companion VeoWatermarkRemover binary.

Quick Start (CLI)

# Single image, automatic detection
GeminiWatermarkTool image.png

# Force processing even if detection fails
GeminiWatermarkTool --force image.png -o clean.png

# Batch folder → output folder
GeminiWatermarkTool -i ./watermarked/ -o ./clean/

# Use legacy profile only
GeminiWatermarkTool --legacy old_image.png

Quick Start (GUI)

  1. Run the executable (no installer needed).
  2. Drag‑and‑drop an image or a folder onto the window.
  3. The tool auto‑detects the watermark; adjust the region or select a different profile if needed.
  4. Press X to apply, V to toggle before/after view, Z to revert.
  5. For video, launch the separate VeoWatermarkRemover binary and drop an MP4 file.

Why It Matters

  • Deterministic restoration: Unlike AI‑inpainting tools that hallucinate content, this method mathematically reverses the exact blending that added the watermark, yielding a faithful reconstruction when the original image hasn’t been altered.
  • Lightweight & offline: A single portable executable with optional GPU‑accelerated denoising, no cloud services required.
  • Up‑to‑date: Supports the latest Gemini 3.5 watermark layout and provides an automatic fallback for older images.
  • Extensible: Exposed via MCP so future AI agents (Claude, Cursor, etc.) can call it directly.

Where to Find More

Bottom Line

GeminiWatermarkTool is a genuine, open‑source utility focused on a very specific AI‑related problem: removing the proprietary watermark that Google’s Gemini model stamps on its outputs. It combines a mathematically exact reverse‑alpha algorithm with optional AI‑based denoising, offers both CLI and a polished GUI, and even provides an API for AI‑agent integration. For anyone needing a reliable, offline way to clean up Gemini‑generated media, this repository is directly on‑topic and ready to use.

Related

  • Project
    wiltodelta/remove-ai-watermarks

    A library and CLI tool to remove visible labels, invisible pixel watermarks, and AI provenance metadata from AI-generated images and videos.

  • Project
    aloshdenny/reverse-SynthID

    A reverse-engineering project that detects and removes Google's invisible SynthID watermarks from Gemini-generated images using spectral analysis and a multi-stage adversarial pipeline.

  • Project
    zuruoke/watermark-removal

    A machine learning project that uses image inpainting to remove watermarks from images, creating seamless results.

  • Project
    guillaumemeyer/watermarks-remover

    A service and agent skill for stripping AI provenance marks and watermarks from text and files across multiple vendors and formats.

  • Dispatch
    Google Gemini AI Image Verification

    Google is integrating SynthID watermarking technology into the Gemini app to allow users to verify if images were generated or edited by Google AI.