Flux Dev2 Pro

Introduction

Flux-Dev2Pro is a fine-tuned version of the Flux-Dev model, optimized to improve LoRA (Low-Rank Adaptation) training results. This model addresses the challenges in LoRA training on Flux-Dev by incorporating guidance distillation, which aligns the LoRA training with the original model's training process.

Architecture

Flux-Dev2Pro builds upon the transformer architecture of Flux-Dev, focusing on enhancing LoRA training through finite tuning. The model undergoes extensive training involving two epochs with 3 million high-quality images to ensure superior performance.

Training

The training process for Flux-Dev2Pro involves fine-tuning the original Flux-Dev model. This process includes training with a large dataset to ensure the model's performance aligns with expectations. The improved model can then effectively apply LoRA, similar to applications in other advanced models like SDXL.

Guide: Running Locally

To run Flux-Dev2Pro locally, you need to use the diffusers library. Below are the steps to get started:

1. Install the diffusers library if you haven't already.

2. Load the model using the following code snippet:

   from diffusers import FluxTransformer2DModel
   
   transformer = FluxTransformer2DModel.from_pretrained("ashen0209/Flux-Dev2Pro", torch_dtype=torch.bfloat16)
   

3. Ensure you have a compatible environment with sufficient computational resources, preferably with a cloud GPU service for optimal performance.

License

The license for Flux-Dev2Pro is not explicitly mentioned in the provided documentation. Please refer to the Hugging Face model page or repository for detailed licensing information.