Crafting a LoRA

Creation, Artistry, Methodology — Part 2 Training Results

The full article can be found here:
https://artfullyprompt.medium.com/crafting-a-lora-51683062174d

Introduction

Creation, Artistry, Methodology

The goal of this two part series is to show the process of creating a LoRA, or low rank adaption for use in text to image generation processing. I did my utmost to include every parameter, setbacks, python scripts etc that took me from the beginning to the end of this article.

I apologize for the length. In all things I strive for transparency and reproduceability. In order to achieve this goal, there are long lists of prompt data, extensive representation of each of the activation phrases, and limited descriptions.

Once you understand the purpose of a section, you can skim until you find what you may be looking for. It is not necessary to read every prompt, or examine every set of three images to fully understand what I have done here.

My inspiration comes solely from a desire to understand more about LoRA generation and a curious nature. My hope is that this article will help you in creating your own LoRAs, provide a point of reference for the different ways that you can adjust a prompt or generation settings, or a place to come and ask questions or provide feedback on your own projects.

The article was split into two articles because of the length of the original was difficult for me to scan back and forth while making edits to the article. My thought was that if it’s difficult for me, its going to be impossible for my audience.

Lastly, the target content of the LoRA is not going to be aesthetically pleasing for everyone. I would urge you to ignore specifically what you are seeing, and instead look at the differences between sets, with a goal of understanding how training data sets, training tags, activation phrases, generation parameters, and testing results all relate to each other. This is the point of the two articles.

Thank you for your time!

The completed LoRA can be found at the link below.

artfully_YOUNGALIEN_SDXL (hosted on civitai.com)

If you are interested in how the training data was created and the parameters that were used in generating the training data, read Part 1 at the link below

Crafting a LoRA | Creation, Artistry, Methodology — Part 1 Training Data and Parameters (medium)