sd 1 5 mse vae

Introduction

This repository features a modified version of Stable Diffusion v1.5, where the original VAE (Variational Autoencoder) is replaced with a new MSE VAE. The updated VAE can be found at stabilityai/sd-vae-ft-mse-original. This project is intended for experimentation, allowing users to explore and test the effects of the new VAE implementation.

Architecture

The primary modification in this repository is the integration of a new MSE VAE into the Stable Diffusion v1.5 framework. The project's architecture centers around swapping the VAE components to potentially enhance image generation, though the changes in results are noted to be minimal.

Training

Details on specific training processes are not provided. The repository encourages experimentation with the new VAE setup to see how it impacts the model's performance.

Guide: Running Locally

1. Clone the Repository: Clone this repository to your local machine.
2. Install Dependencies: Ensure you have the necessary libraries, such as those required for running Stable Diffusion models.
3. Configure the Model: Swap the original VAE with the new MSE VAE using the provided resources.
4. Test and Experiment: Run the model to generate images and observe the differences.

For improved performance and faster computation, consider using cloud GPUs from providers like AWS, Google Cloud, or Azure.

License

Usage of this model requires acceptance of the CreativeML Open RAIL-M license. For more details, refer to the license file.