Dreambooth Training SDXL Using Kohya_SS On Vast.ai

Dreambooth Training on Stable Diffusion XL Using Kohya_SS’s Scripts on Vast.ai

If you’re not a Medium subscriber, you can read this post for free here.

Table of Contents

Introduction
Pre-requisites
Vast.ai
Jupyter Notebook
Using Captions
Config-Based Training
Aspect Ratio / Resolution Bucketing
Resume Training

Updated : 4th December 2023

Introduction

Stability AI released SDXL model 1.0 in July 2023. It is the successor to the popular v1.5 model and the somewhat less popular v2.x models. It is a much larger model compared to its predecessors in terms of number of parameters and the default resolution is 1024x1024. and you can readd more about it here.

SDXL comes with two models : the base and the refiner. Basically the base model produces the raw image and the refiner (which is an optional pass) adds finer details.

Training a subject on top of SDXL will be somewhat familiar if you have done it before regardless of repo. In this guide…