Can we train a person and style in your extension in one model and both staying intact?

[wcarletsdrive]

If so:
How many reg images are recommended for training sets like a formula? Is it 10 reg images per training image?

Do we train with train text encoder to 50%? Or completely off/on?

For the json, do we keep the same format as such:
[
{
"instance_prompt": "xxxxxx",
"class_prompt": "per",
"instance_data_dir": "C:/Users/user/foldername",
"class_data_dir": "C:/Users/user/Classes/foldername"
}
{
"instance_prompt": "xxxxx",
"class_prompt": "wom",
"instance_data_dir": "C:/Users/user/foldername",
"class_data_dir": "C:/Users/user/Classes/foldername"
}
]

I trained with shivam's repo, his november 3rd commit since that was working for me, and training is fine separately, but when I do multi training with the same datasets together of a person and style, I lose the style a lot and the person is kind of the same. Idk why this happens.