THEANTLERS LORA GUIDE

I have no knowledge on model creation outside of whats here . I downloaded the settings somewhere and lightly modified them, so im quite ignorant, regardless, the results speak for themselves and i find most of my loras to be adequate

DATASET

I use 30-50 images, always going for consistency in style and latest images, I tag them with https://github.com/Particle1904/DatasetHelpers/releases
Go to "generate tags", i use wdv3 model and default settings. After tagging, on "proccess tags" i add "NAMEOFTHEARTIST" on tags to add and tags to emphasize

TRAINING

For the training itself I use https://github.com/derrian-distro/LoRA_Easy_Training_Scripts you can find an installation tutorial on the github page, these are my settings:

SETTINGS FOR PONYXL LORAS: https://files.catbox.moe/k5xrqt.toml
SETTINGS FOR PONYXL DORAS: https://files.catbox.moe/ocvjsv.toml
SETTINGS FOR ILLUSTRIOUS LORAS: https://files.catbox.moe/94o4i3.toml (BEST RESULTS)
SETTINGS FOR CIVITAI TRAINER (Worse results but i have buzz to spare- try to get 2000-2300 steps): https://imgur.com/a/VmjULji
All of these settings are used for style training - if you want to train a character, I use these: https://files.catbox.moe/calc5r.toml

If you go with derrian's lora trainer:
On subset args, simply add the path of your folder with the images and captions. Make sure shuffle tags is on!

Make sure to select the correct model and vae paths, and then we add the correct output folder, the name we want for the model and we start training, that is all

Thanks to LUNA for introducing me to derrian's kohya implementation! Great fella, learned tons from him!