DermaFlux

This repository contains LoRA weights and requires the Flux.1 base model. Use of this model is subject to the license of the base model.

Model Description

DermaFlux LoRA is a parameter-efficient fine-tuning of the Flux.1 text-to-image model designed to generate clinically grounded skin lesion images from structured natural language prompts. The model enables synthesis of dermatology images conditioned on clinically relevant attributes such as:

• lesion asymmetry
• border irregularity
• color variation

The goal of the model is to augment dermatology datasets with realistic synthetic samples to improve downstream tasks such as skin lesion classification. The LoRA weights modify the base Flux.1 generative model without requiring full model retraining.

Installation

The environment setup and inference scripts are provided in the official repository:

https://github.com/dermaflux/dermaflux

Please follow the installation instructions in the repository to create the required environment and install the dependencies.

The repository includes:

• environment setup instructions
• required Python packages
• inference script for running the model

After setting up the environment, download the LoRA weights from this Hugging Face repository and use them with the inference code provided in the GitHub repository.

Citation

If you use this model, please cite the DermaFlux paper:

@misc{galanakis2026dermafluxsyntheticskinlesion,
      title={DermaFlux: Synthetic Skin Lesion Generation with Rectified Flows for Enhanced Image Classification}, 
      author={Stathis Galanakis and Alexandros Koliousis and Stefanos Zafeiriou},
      year={2026},
      eprint={2603.16392},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.16392}, 
}