SORA 2 is a State-of-the-art model by OpenAI and for the past few days, being on platforms like Instagram and Twitter, I've noticed how many non-technical people just assume the video is real despite the watermark.
Think what would happen if there was no watermark? This is the reason that this project exists. It's not to abuse the great initiative by OpenAI to put logos onto every generation (though temporarily there's also an easy way to bypass that which I wouldn't cover), it's to hopefully encourage them to be harsher and more obvious with it in some form.
Sweeta is an AI-powered watermark removal tool specifically designed for SORA 2 video generations. It Uses advanced inpainting models (LaMA) and intelligent detection algorithms, it can seamlessly remove watermarks while (mostly) preserving the original image quality.
- OS: Windows 11, macOS 12+, or Ubuntu 20.04+
- RAM: 16GB or more
- Storage: 10GB free space
- GPU: NVIDIA GPU with 4GB+ VRAM (or Apple Silicon for macOS)
- CPU: Multi-core processor (Intel i5/AMD Ryzen 5/Apple M1 or better)
- Python & Conda: Install Miniconda (recommended) or Anaconda
-
Open Command Prompt or PowerShell as administrator
-
Navigate to the project folder
-
Run the installation script:
cd path\to\Sweeta windows\install_windows.batOr for PowerShell:
powershell -ExecutionPolicy Bypass -File windows\install_windows.ps1
-
Follow the on-screen instructions
# Create the conda environment
conda env create -f environment.yml
# Activate the environment
conda activate py312aiwatermark
# Install additional dependencies (including required version pins)
pip install PyQt6 transformers iopaint opencv-python-headless "pydantic<2.0.0"
# Download the LaMA model
iopaint download --model lamaAlternative: Using requirements.txt
conda create -n py312aiwatermark python=3.12
conda activate py312aiwatermark
pip install -r requirements.txt
iopaint download --model lama# Navigate to the project directory
cd /path/to/Sweeta
# Run the setup script
bash unix/setup.sh
# Or manually:
conda env create -f environment.yml
conda activate py312aiwatermark
pip install PyQt6 transformers iopaint opencv-python-headless "pydantic<2.0.0"
iopaint download --model lamaAlternative: Using requirements.txt
conda create -n py312aiwatermark python=3.12
conda activate py312aiwatermark
pip install -r requirements.txt
iopaint download --model lamaAccess the Colab notebook from here and follow the instructions.
-
Activate the conda environment:
conda activate py312aiwatermark
-
Launch the GUI application:
python remwmgui.py
(would be happy to prepare a hugging face port for Spaces too, which would technically be better but would require community GPU access)
conda activate py312aiwatermark
python remwm.py <input_path> <output_path> [options]Example:
python remwm.py input_video.mp4 output_video.mp4 --max-bbox-percent 15 --force-format MP4 --transparent --overwriteAvailable options:
--max-bbox-percent: Detection sensitivity (default: 10.0)--force-format: Output format (PNG, WEBP, JPG, MP4, AVI)--transparent: Make watermark areas transparent--overwrite: Overwrite existing files
Refer #ui.yml.example
Configuration Options
- Input Path: Select your source file or folder
- Output Path: Choose where to save processed files
- Overwrite Files: Enable to replace existing output files
- Transparent Watermarks: Make watermark areas transparent (PNG only)
- Max BBox Percent: Adjust detection sensitivity (1-100%)
- Output Format: Choose PNG, WEBP, JPG, or keep original format
For detailed troubleshooting steps, see TROUBLESHOOTING.md.
Solution: This is a Pydantic version incompatibility issue. The installation scripts now automatically install the correct version. If you installed manually, run:
pip install "pydantic<2.0.0"Solution: This is a version compatibility issue. The installation scripts now automatically install the correct version. If you installed manually, run:
pip install "huggingface-hub<0.20"
pip install --upgrade iopaintSolution: Install PyTorch with CUDA support. First, uninstall existing PyTorch:
pip uninstall torch torchvision torchaudioThen reinstall with CUDA support (for CUDA 11.8):
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118Or for CUDA 12.1:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121Verify CUDA is available:
python -c "import torch; print(f'CUDA available: {torch.cuda.is_available()}')"Solution: Ensure Conda is properly installed and added to your system PATH environment variable.
Solution: Try installing dependencies individually:
pip install PyQt6
pip install transformers
pip install iopaint
pip install opencv-python-headless
pip install "pydantic<2.0.0"Solution: Verify the environment is activated:
conda activate py312aiwatermark
python --version # Should show Python 3.12.xSolution: Ensure stable internet connection and retry:
iopaint download --model lamaSolution: Install PyTorch with CUDA support:
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118If you run into any issues or have something to say, reach out! I'd be happy to talk :) Twitter, LinkedIn
I (tired to) microblog the development process in #journal.md but uh, read it at your own risk lol Will put out a blog or something similiar too.
This project is licensed under the Apache License 2.0 - see the LICENSE file for details. Thanks to D-Ogi for the WatermarkRemover-AI model which was heavily modified for this project.
THIS SOFTWARE IS PROVIDED FOR EDUCATIONAL AND RESEARCH PURPOSES ONLY. Use this tool responsibly and ethically.
Made with <3 by Kuber Mehta
Star this repo if you found it cool