WatermarkRemover-AI: Open-Source Image & Video Watermark Remover

WatermarkRemover-AI is an open-source project for detecting and removing visible watermarks from images and videos.

It uses Florence-2 to locate watermark regions and LaMA to reconstruct the covered pixels. You can run it through a PyWebview desktop app or a Python CLI.

The Watermark removal tool is designed for local workflows that need batch image processing, detection previews, or video controls for recurring and fading overlays.

Your files stay on the workstation during processing. Video exports retain the source audio track when FFmpeg is installed.

View WatermarkRemover-AI on GitHub

Features

Florence-2 detects visible watermark regions from the default watermark prompt or a custom detection prompt.
LaMA inpaints accepted watermark regions from surrounding visual context.
Preview Mode shows accepted and rejected detection boxes before removal starts.
Batch mode processes a folder of images in one job.
Video processing supports detection skipping plus fade-in and fade-out mask controls.
Image output can use LaMA inpainting or transparent detected regions.

See it in Action

How Watermark Detection and Removal Work

Florence-2 receives a text prompt and returns bounding boxes for matching regions. WatermarkRemover-AI checks each box against max-bbox-percent before removal. The default threshold accepts boxes that cover no more than 10% of the image.

Accepted boxes become masks for LaMA, which reconstructs the masked area from nearby image content. For images, the transparent option replaces accepted regions with transparency.

Preview Mode stops after detection and shows which boxes pass the size threshold. If the detector misses a watermark, use a more specific detection prompt. If the correct box is rejected because it exceeds the threshold, raise max-bbox-percent.

How to Install WatermarkRemover-AI

Windows

Clone WatermarkRemover-AI and run the PowerShell setup script:

git clone https://github.com/D-Ogi/WatermarkRemover-AI.git
cd WatermarkRemover-AI
.\setup.ps1

Launch the desktop UI with run.bat after setup finishes.

Linux and macOS

Linux and macOS require Python 3.10 or newer. Clone WatermarkRemover-AI, make the setup script executable, and run it:

git clone https://github.com/D-Ogi/WatermarkRemover-AI.git
cd WatermarkRemover-AI
chmod +x setup.sh
./setup.sh

Start the desktop UI with:

./run.sh

FFmpeg for Video Audio

FFmpeg is required to retain the source audio track in processed videos. Install it with sudo apt install ffmpeg on Debian-based Linux systems or brew install ffmpeg on macOS. On Windows, install FFmpeg and add it to the system PATH.

How to Use the GUI

Launch run.bat on Windows or ./run.sh on Linux/macOS.
Select Single File or Batch mode.
Choose the input and output paths.
Use Preview Mode to inspect the detected watermark regions before removal.
Adjust the detection prompt, bounding-box threshold, output settings, or video controls when needed.
Start processing after the preview targets the intended region.

How to Use the CLI

Process one image with an input path and output directory:

python remwm.py input.png output_folder/

Run a batch job with overwrite enabled, a larger detection threshold, and PNG output:

python remwm.py ./images ./output --overwrite --max-bbox-percent=15 --force-format=PNG

Inspect detection boxes without running the removal stage:

python remwm.py input.png --preview

For video, this example runs detection every third frame and extends the mask by half a second before and after detected frames:

python remwm.py video.mp4 ./output --detection-skip=3 --fade-in=0.5 --fade-out=0.5

CLI Options

Option	Behavior
`--overwrite`	Replace existing output files.
`--transparent`	Make detected regions transparent for image output.
`--max-bbox-percent`	Set the maximum accepted detection size; default 10%.
`--force-format`	Force PNG, WEBP, JPG, MP4, or AVI output.
`--detection-prompt`	Replace the default `watermark` detection prompt.
`--detection-skip`	Run video detection every N frames; accepted range 1–10.
`--fade-in`	Extend a video mask backward by N seconds.
`--fade-out`	Extend a video mask forward by N seconds.
`--preview`	Return detections without running removal.

Pros

Runs locally after setup.
Both a desktop GUI and a CLI for repeatable jobs.
Preview Mode exposes detection problems before inpainting starts.
Supports image folders plus video-specific skip and fade controls.

Cons

Requires local installation and model downloads.
Video audio requires FFmpeg.
Ambiguous overlays increase the risk of missed or incorrect detections.
Complex textures and large masks are more likely to show visible artifacts.

Alternatives & Related Resources

FAQs

Q: Does WatermarkRemover-AI require an account or API key?
A: No. The local workflow does not require an account or API key. Initial setup downloads dependencies and model files, then the application processes media on your machine.

Q: Does Preview Mode modify the source file?
A: No. Preview Mode runs detection and returns the detected regions without starting the LaMA removal stage.

Q: Does removing a watermark change the media license?
A: No. Removing a visible watermark does not change copyright ownership or license terms. Process only media you own or have permission to modify.

Last Updated: Aug 10, 2026