A NEW SERIES OF MODELS TO COME!
This is the first SDXL model I've successfully trained and even at that it isn't perfect.
Works Best at a weighting of 0.9 - 1.0.
Generation parameters: suzu, fox ears, fox tail, brown eyes, orange-brown hair, yellow miko kimono, medium hair, ahoge, hair between eyes
Uses the same dataset from Suzu 1.5 v4
Additional Details / Training Info (For anyone wanting to train SDXL for themselves)
Trained using OneTrainer (Link) (Preset File)
20 Epochs (Final model uses the full 20)
10 Dataset Repeats
0.0003 Learning rate
896x896 Train Resolution (Could go higher)
(NOTE: FOR ANYONE LOOKING TO TRAIN ON THE KOHYA SDXL COLAB, DO NOT. IT IS BEYOND BROKEN AND ANY MODEL IT "TRAINS" DOESN'T WORK)