CLoRA: Contrastive Test-Time Composition of Multiple LoRA Models for Image Generation

This repository contains the implementation of CLoRA, a method for composing multiple LoRA (Low-Rank Adaptation) models at test time using contrastive learning for improved image generation with Stable Diffusion.

UV is a fast Python package installer and resolver that ensures consistent dependency resolution.

If you don't have UV installed:

# Install UV using the official installer
curl -LsSf https://astral.sh/uv/install.sh | sh

# Or using pip
pip install uv

# Or using pipx
pipx install uv

# Clone the repository
git clone https://github.com/gemlab-vt/clora
cd CLoRA

# Create virtual environment and install dependencies
uv sync

# Check if everything is installed correctly
uv run python -c "import torch; print(f'PyTorch version: {torch.__version__}'); print(f'CUDA available: {torch.cuda.is_available()}')"

The main demonstration and examples are provided in the Jupyter notebook.

# Start Jupyter Notebook using UV
uv run jupyter notebook notebook.ipynb

# Or start Jupyter Lab
uv run jupyter lab notebook.ipynb

If you use this code in your research, please cite:

@InProceedings{Meral_2025_ICCV,
    author    = {Meral, Tuna Han Salih and Simsar, Enis and Tombari, Federico and Yanardag, Pinar},
    title     = {Contrastive Test-Time Composition of Multiple LoRA Models for Image Generation},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2025},
    pages     = {18090-18100}
}

This project is licensed under the MIT License - see the LICENSE file for details.

• Author: Tuna Han Salih Meral
• Email: [email protected]

For questions about the implementation or research, please open an issue or contact the author directly.