This is not Normal Checkpoint Model (Tested and Experimented) and It is hard to use for you, you would get many problems with this my model.
You should carefully prompt, it is very hard to control, my recommends are you should use it with Right LoRA, or else you would un-normal results
Model Info-------------------------------------------------
Base Model: SD.1.5
Model used to train: Beautiful Realistic Asia or BRA V5 (Tested)
Trained at Resolution 1024.
Learning Rate 1e-6, Max Steps at 3500.
Optimization AdamW8Bit, LR Scheduler Cosine with Restart.
VAE (Used to Train): None, Sampling Method: K_DPM_2
Clip Skip: 2 (You should use Clip Skip 2)
[New NOTE: Clip Skip 2 is not really meant to be good as my recommend because my Model is making many different to Clip Skip 1-5 - So No MEANS to any Clip Skip 1-5]
Number of Images used to train is about 100 Images (All not Normal Images)
Activation Word: HighHumanion.
You can use this word (or Activation Word) in your prompt as following syntax: HighHumanion:<weight value herer>,
and you would get nothing special thing, you can use this word or not, it will impact your prompt but as I said, I can't predict what happen!
NOTE. Above Images is from SeaArt, you should use Quality is High Quality, but It would be not special, it is likely BRA V5, with High Quality, you would not too care about many problems.
For Quality is Ultra High Quality in SeaArt, it is very hard to predict Results, it can surprising or painful to you, so you should be carefully about Ultra High Quality.
Guide: Don't try to use 'HighHumanion:<weight value herer>' in your prompt it can be hard to use and predict any result; You shouldn't use word 'restore' such as restore face, it is hard to predict. However, it is hard to predict from any words in your prompt with my model, so don't too care about my this guide, it is still un-predictable
Hope you will enjoy it, So No Pain No Gain
You can post review posts with your impressive results, it is helpful to my Model