Open F L U X.1

Introduction

OpenFLUX.1 is a fine-tuned version of the FLUX.1-schnell model, designed to remove distillation and enable fine-tuning. FLUX.1-schnell, licensed under Apache 2.0, originally did not support fine-tuning. OpenFLUX.1 seeks to provide an open-source, permissively-licensed alternative that can generate high-quality images in 1-4 steps.

Architecture

The model utilizes a pipeline that differs from the original FLUX.1 schnell. This is due to the removal of distillation, which allows the use of classic Classifier-Free Guidance (CFG). The necessary pipeline for this model can be found in the open_flux_pipeline.py file within the repository.

Training

OpenFLUX.1 has been trained to remove the distillation characteristic of the original FLUX.1-schnell model. The creator notes that the model is in its beta phase (V0.1.0) and further updates are expected. Users are encouraged to experiment with and fine-tune the model, although ongoing updates may occur.

Guide: Running Locally

1. Clone the Repository: Obtain the OpenFLUX.1 repository from Hugging Face.
2. Install Requirements: Ensure you have Python and dependencies installed. Use pip install -r requirements.txt if a requirements file is available.
3. Modify Pipeline: Use the open_flux_pipeline.py for the model setup.
4. Run the Model: Configure the CFG to 3.5 for optimal performance.
5. Cloud GPUs: Consider using cloud GPU services like AWS, GCP, or Azure for efficient model training and inference.

License

OpenFLUX.1 is licensed under the Apache 2.0 license, allowing for open-source use and modification.