👋 Introduction
This guide provides an overview of my training process for LoRas using my model, Pony Realism for Kohya_ss.
Training a LoRa model involves several steps, from preparing your dataset to fine-tuning the model for optimal performance. While this guide offers a brief walkthrough of my process, it will not delve deeply into all settings, as I am sharing my preset. Some values will need to be adjusted based on your hardware and data, such as the number of epochs and batch size.
📦 Preparations
First, you need some indispensable items:
A collection of images for your character/style: Quality is more important than quantity. I usually select 14-50 high-quality images.
Captions: You can generate these using Kohya_ss Utilities -> Captioning. I mostly use WD14 Captioning, but in some situations, BLIP can be better. I add a unique prefix and run the default settings.
📝 Steps Calculation
The calculation for the epochs/step that I do is:
number of images * steps = ~900
So for example in the following dataset:23 images * 40 = 920.
That 40 value goes to the folder as number_subject:
Lastly to calculate the epochs I do:
steps * epoch / batch size = ~2500
This gives me a epochs of 6:
920 * 6 / 2 = 2760
📜 Training Preset
The preset uploaded here is what I use on everything, character/style/concept, it should be ready to go, you could need to change the batch size and epochs based on your hardware, mine for reference is the following:
RTX 4090
i7 13700k
64 Gb Ram 6400 MHz