masc_lora

Introduction

MASC LoRA is a project by Divergent AI, developed in collaboration with Gen Hunks. It aims to address the bias towards female images in AI generation by providing photorealistic and diverse male representations. The model is based on Flux Schnell and is designed to produce artistic and editorial-quality results.

Architecture

The architecture of MASC LoRA is built on the Flux Schnell framework. It utilizes a training dataset of 400 curated images with diverse male representations, focusing on high-quality, photorealistic outputs. The model incorporates comprehensive captioning through Joy Caption and ComfyUI, further enhanced by LLMs for accurate visual content understanding.

Training

MASC LoRA is trained with a diverse dataset, capturing various skin tones, body types, and artistic styles. The training process involves 40-80 GPU hours, focusing on photorealism and diversity. Despite being an early prototype (V0.3), it effectively generates lifelike male imagery. Ongoing improvements aim to expand the dataset, enhance quality control, and refine image captioning.

Guide: Running Locally

1. Setup Environment: Ensure you have Python installed along with necessary libraries like PyTorch and Hugging Face Transformers.
2. Download Model Weights: Obtain the model weights in Safetensors format from the provided link.
3. Install Dependencies: Use package managers like pip to install any additional dependencies required by the model.
4. Run Inference: Use a script or Jupyter Notebook to load the model and run inference on your chosen input prompts.
5. Cloud GPUs: Consider using cloud services like AWS, Google Cloud, or Azure for accessing GPUs, which can significantly speed up the inference process.

License

MASC LoRA is licensed under the Apache 2.0 License, allowing for both personal and commercial use, with the condition of providing proper attribution.