Asian Girl Face:Training based on synthetic data.

The face of a beautiful Asian girl,she is a completely fictional character. At first, she was just some of my attempts on SD1.5 version. She originated from some results I got when I used role consistency models such as ip-adpter. I think these results are different from real reference pictures but at the same time have some beautiful facial features that we can consider as common consensus. I chose some faces that I approve of and started to establish the first version.

She may think of some characters, but in fact, this is all based on some character features that I think meet my aesthetic preferences and at the same time have some recognizable features. Starting from one or two reference pictures that meet the needs (also generated by AI), and then through a series of effective image generation and high-definition magnification based on role consistency, I have obtained a model training data set of hundreds of images. Then, this is what you see now. A beautiful girlfriend who won't lose her temper and is infinitely close to a real person.

It should be noted that due to the limitations of the Flux.1 model, it may not be so easy for you to achieve NSFW. But perhaps the beauty lies in packaging, doesn't it?

LoRA model weight value: 0.8 - 1.2.

Base model: Flux.1 Dev fp8.

Step: 30 - 50.

If the Shenell model is also used, Step: 4 - 8 will be good.

Enjoy it.