Stable Diffusion - How to Train LoRA for Style - LoRA Part 1

Learn how to train LoRA (Low-Rank Adaptation) for style in Stable Diffusion through this comprehensive tutorial video. Explore the process of creating custom style models using both local GUI and Google Colab methods. Discover techniques for dataset preparation, including batch renaming, cropping, and captioning. Master the use of kohya_ss GUI for project setup and training parameter configuration. Gain insights into testing and evaluating LoRA models using AUTOMATIC1111 webui and x/y/z plotting. Delve into anime-style LoRA creation, utilizing tools like the AUTOMATIC1111 webui extension tagger and modifying Danbooru-style tags. Troubleshoot common errors and learn to combine character and style LoRAs for enhanced results in AI-generated artwork.

Intro
Why is LoRA a new standard?
Required resources
Overview of video parts 1. local GUI and 2. Google Colab Notebook
Part 1. local GUI training
PC hardware specs
Dataset preparation
Batch renaming filetypes to a new filetype
Batch renaming attempt #2
BIRME cropping
BLIP captioning
How to write good captions
Captioning example #2
Creating project folder structure using kohya_ss GUI
Training parameters
Why the training steps in the console are different from max train steps
Console output after training commences
What you get after training finishes
Testing the style LoRA in AUTOMATIC1111 webui
Using x/y/z plot in scripts to vary parameters
Checking x/y/z plot in photoshop
Testing the final epoch LoRA
Verifying LoRA's flexibility with prompting
Part 2. Anime style LoRA
Dataset preparation on redjuice/shiru artwork
Using AUTOMATIC1111 webui extension tagger
Modifying Danbooru style tags
Using the kohya LoRA DreamBooth colab notebook
Fixing the error that occurred
Preliminary testing of redjuice style LoRA
Checking x/y/z plot in photoshop
Combining character and style LoRAs