It was done using civitai lora, so there is a limit to the training, for best results use the same prompt that I used in my images

I use lora at <lora:zoe_monroe:0.5>