Suggested 8 steps and cfg 1 for generation with this Lora.
==========================================
Beta3 reduce the Learning rate to 8e-5, better result.
==========================================
Beta2 reduce the Learning rate to 9e-5, better facial expression.
==========================================
This Lora is a valid resource intended for use in content generation, and open to public for review.
Codename "KW" is an artificial character that was created in research lab. The dataset was heavily edited in multiple software and not intended to imitate any person in real.
Description
The AI-Toolkit settings are as follows:
Optimizer: AdamW
Training Precision: fp32
Save Precision: float32
EPS: 1e-8
Learning rate: 9e-5
Linear: 256
Linear_alpha: 256
Resolution: 1024x1536
Batch Size: 1
training_seed: 1234
transformer_only: false
caption_dropout_rate: 0.0
ema removed
quantize: false
quantize_te: false