v5.0a

  "unetLR": 0.0005,
  "clipSkip": 2,
  "networkDim": 128,
  "numRepeats": 10,
  "resolution": 1024,
  "lrScheduler": "cosine_with_restarts",
  "noiseOffset": 0.1,
  "networkAlpha": 64,
  "optimizerType": "AdamW8Bit",
  "textEncoderLR": 0.00005,

bigger dataset 

v3.0b:

Text Encoder learning rate:0.000005

Unet learning rate:0.0001

LR Scheduler:cosine

Optimizer:AdamW8bit

Network Dim:64

Network Alpha:32

v3.0a:

Bigger training dataset

0.8 have best result

1 look weird for me

Text Encoder learning rate:0.00001

Unet learning rate:0.0001

LR Scheduler:constant

Optimizer:AdamW8bit

Network Dim:32

Network Alpha:16