SD-Elysium-Model

Introduction

The SD-Elysium-Model is a versatile AI model designed for generating high-quality, realistic and anime-style images. It includes two variants: the general Elysium model, which focuses on realistic styles, and the Elysium Anime model, which is tailored for detailed and versatile anime-style images.

Architecture

General Model: Utilizes a VAE (Variational AutoEncoder) setup with vae-ft-mse-840000 from StabilityAI. The recommended sampler is DPM++ 2M Karras with a Clip skip set to 1.
Anime Model: Employs kl-f8-anime2.ckpt VAE from Waifu Diffusion. It supports Clip skip settings of 1 or 2 and uses the same sampler, DPM++ 2M Karras.

Both versions are optimized to work effectively with booru tags for enhanced image generation quality.

Training

The model's training details such as dataset size, epochs, or specific frameworks used are not provided. However, it leverages existing VAE architectures and samplers suited for its respective styles.

Guide: Running Locally

To run the SD-Elysium-Model locally, follow these basic steps:

Set Up Environment:
- Ensure you have Python installed.
- Install any required dependencies or libraries typically via pip.
Download the Model:
- Access the model files from Hugging Face: Elysium.
Configure the Model:
- Load the appropriate VAE and configure the sampler settings as specified (DPM++ 2M Karras).
Run the Model:
- Execute the model using your preferred interface or script, ensuring you apply the recommended negative prompts to optimize the output quality.

For enhanced performance, consider using cloud GPU services such as AWS EC2, Google Cloud, or Azure, which can provide the necessary computational resources.

License

The SD-Elysium-Model is released under the OpenRAIL license, allowing for wide usage while ensuring compliance with specified terms and conditions.