WatermarkRemover

What it does: A tool that removes fixed watermarks from videos in batches, powered by the LAMA model.

Requirements

• Python 3.10 (exact version required)

Installation

1. Clone the repository and enter the folder.
2. (Optional but recommended) Create and activate a Python virtual environment.
3. Install base dependencies with pip install -r requirements.txt.
4. Install PyTorch — choose one of two options:
   • CPU version: pip install torch
   • GPU version (needs NVIDIA graphics card):
     • Install CUDA Toolkit from NVIDIA's download page.
     • Install cuDNN matching your CUDA version.
     • Install the GPU build of PyTorch using the command from PyTorch's official site (e.g., pip3 install torch==2.6.0+cu126 ...).

The program automatically detects whether a GPU is available and picks the processing mode accordingly.

Usage

Run from the command line:

python watermark_remover.py --input /path/to/videos --output /path/to/output

Add --preview to see the effect before processing.

Command-line options

Parameter	Short	Description	Default
--input	-i	Folder containing video files	. (current folder)
--output	-o	Folder for processed videos	output
--preview	-p	Show effect preview before processing	Disabled

How it works

1. Select the watermark area: The program shows one frame of the video. You manually draw a box around the watermark, then press SPACE or ENTER to continue.
2. Preview (optional): If enabled, shows the processed result. Press SPACE or ENTER to confirm, or ESC to cancel and exit.
3. Process the video: The first run downloads the LAMA model, which takes a while.
4. Output: Results are saved as MP4 files.

Limitations

• Only works on watermarks that stay in a fixed position — moving watermarks are not supported.
• All videos processed in one batch must have the same dimensions.
• All videos in one batch must have the same watermark.

Troubleshooting GPU issues

If you see No GPU detected, using CPU for processing, the README suggests:

• Check that Python is 3.10.
• Verify you installed the GPU version of PyTorch, not the CPU one (the README quotes Lama Cleaner's docs: if the tool isn't using the GPU, a CPU-only PyTorch is likely installed).
• Make sure your CUDA, cuDNN, and PyTorch versions are compatible with your graphics card:
  • Check NVIDIA's CUDA compatibility list for your GPU.
  • Pick a matching cuDNN version from NVIDIA's archive.
  • Pick the matching PyTorch build from PyTorch's site.

When the GPU is detected correctly, the program prints something like GPU detected: NVIDIA XXX Using GPU for processing.

Star history

The README includes a star-history chart showing the project's GitHub stars over time.