Veo Watermark Remover

NEW: Watch the demo video on YouTube — Veo Watermark Remover in Action

[!WARNING] This is a DEMO/PREVIEW build — CLI only, unsigned executable, early testing phase. See Demo Build Notice for details and known limitations. Windows/macOS may block the executable on first run — see First Run — OS Security Prompts below.

This is a demo/preview build of the upcoming Veo watermark removal feature for GeminiWatermarkTool. Once stabilized, this functionality will be merged into the main GWT release with full GUI support.

Remove the "Veo" text watermark from Google Veo-generated videos using mathematically precise reverse alpha blending — the same proven technique behind GeminiWatermarkTool.

No cloud services. No AI hallucination. No quality loss. Just math.

What's New (v0.4.0)

• macOS stability fix — addresses the SIGBUS crash on Apple Silicon reported in upstream GeminiWatermarkTool #30 / #31. The Vulkan loader is probed before NCNN's GPU init; if it's absent (the default on macOS without MoltenVK), AI denoise transparently falls back to CPU instead of crashing
• --snap-min-size — finer snap-search control on the image-watermark side (default 16, range 8-64), mirroring the existing --snap-max-size flag
• End-to-end CI smoke test — every CI build now runs --denoise ai through a real inference pipeline (Windows / Linux / macOS) before publishing. The previous --version-only check never touched the AI denoise init path, which is how the macOS crash slipped past CI in v0.3.0

Previous Release Highlights (v0.3.0)

• Improved watermark masks — remastered 720p and 1080p alpha maps using golden frame-differencing from multiple video sources (10+ transition pairs)
• Better output quality — upgraded video encoder (High profile, B-frames, multi-reference)
• Enhanced AI cleanup — tuned per-resolution denoise parameters with expanded inference context
• Progress bar — real-time ASCII progress indicator during video processing
• Quieter output — reduced log verbosity in normal mode; use --verbose for detailed diagnostics

How It Works

Just drag and drop. Drop your Veo video onto the executable — it automatically removes the watermark and outputs video_processed.mp4 with audio preserved.

Before / After

Quick Start

Drag & Drop (Windows — Easiest)

1. Download GeminiWatermarkTool-Video.exe (see Download below)
2. Drag your .mp4 file onto the exe
3. Done — output appears as yourfile_processed.mp4 in the same folder

Command Line (All Platforms)

# Simplest — auto-detect, auto-output
./GeminiWatermarkTool-Video video.mp4
Specify output path
./GeminiWatermarkTool-Video -i video.mp4 -o cleaned.mp4
Explicit Veo mode (same result)
./GeminiWatermarkTool-Video --veo -i video.mp4 -o cleaned.mp4

Features

• Single executable — standalone, zero dependencies, zero installation
• Cross-platform — Windows, Linux, and macOS (Universal Binary)
• Direct mp4-to-mp4 — no intermediate files, no external tools needed
• Audio preserved — original audio track is kept without re-encoding
• AI denoise — GPU-accelerated FDnCNN cleanup for residual artifacts (Vulkan)
• Fast — 720p at ~50 fps, 1080p at ~18 fps (Release build, Ryzen 9 + RTX)
• Landscape + Portrait — supports both orientations at 720p and 1080p
• Progress bar — real-time processing progress in the terminal

Technical Details

This tool uses the same reverse alpha blending algorithm as GeminiWatermarkTool:

original = (watermarked - alpha * logo_value) / (1 - alpha)

The Veo watermark alpha maps were extracted using frame differencing from Veo-generated videos with watermark on/off transitions. Multiple golden pairs (Lightning flash, Light bulb, Sunrise scenes) were averaged for maximum accuracy. AI denoise (FDnCNN, NCNN + Vulkan GPU) handles residual edge artifacts from video compression.

For the full technical background on reverse alpha blending, see: Removing Gemini AI Watermarks: A Deep Dive into Reverse Alpha Blending

System Requirements

Note: AI denoise prefers a Vulkan GPU but transparently falls back to CPU when no Vulkan loader is available.

macOS users: macOS does not ship Vulkan natively — install the LunarG Vulkan SDK (which bundles MoltenVK) for GPU acceleration. Otherwise the tool runs in CPU mode by default; this is fully supported but slower than Vulkan inference.

Download

Single file download — no installer, no setup.

Download the latest release from the Releases page.

Verification

SHA256 checksums are provided on each release page. Verify with:

# Windows (PowerShell)
Get-FileHash .\GeminiWatermarkTool-Video.exe -Algorithm SHA256
Linux / macOS
sha256sum GeminiWatermarkTool-Video

If you're not comfortable running an unsigned executable, that's completely fine. This is a demo build for early testing. The Veo removal feature will be integrated into the main GeminiWatermarkTool release when ready, with full source code, CI/CD builds, and cross-platform support.

First Run — OS Security Prompts

Downloaded binaries are not code-signed, so your OS may show a security warning on first launch. This is normal for open-source software distributed outside app stores.

Unblock-File .\GeminiWatermarkTool-Video.exe

xattr -dr com.apple.quarantine GeminiWatermarkTool-Video
chmod +x GeminiWatermarkTool-Video

chmod +x GeminiWatermarkTool-Video
./GeminiWatermarkTool-Video video.mp4

Demo Build Notice

This is a preview/demo build with the following limitations:

• CLI only — GUI support is planned for the main GWT release
• Video only — this build processes video files (.mp4/.mkv/.mov); for Gemini image watermarks, use the main GeminiWatermarkTool

What's Next

When the Veo feature is stable, it will be merged into GeminiWatermarkTool as a unified release supporting both Gemini image and Veo video watermark removal.

About GeminiWatermarkTool

GeminiWatermarkTool is an open-source tool for removing Gemini AI visible watermarks from images. It supports GUI + CLI dual-mode operation, cross-platform deployment (Windows / Linux / macOS / Android), and optional AI denoise with NCNN + Vulkan GPU acceleration.

Author: Allen Kuo (@allenk)

License

This demo build is based on GeminiWatermarkTool (MIT License).

It includes third-party open-source components (NCNN, OpenCV, etc.) under their respective permissive licenses. Full license texts are available in the respective project repositories.