!!! I created this LoRa, but I don't want to be responsible for anything, so be careful when using and distributing it !!!

This LoRa generates a real person known by the pseudonym Sveta Sollar on the internet... I don't want to limit you to generation how to generate images, so I didn't set up this LoRa page on this site as a real person.

Below is an image with all the LoRa training settings and my recommendations for generation. This is my first time creating such a LoRa, so don't judge it harshly.

Weight: 1.0 - 1.05
The more realistic the checkpoint you use, the higher you can raise the CFG Scale. In my examples, this can be clearly seen, but it is better not to raise it above 8,0. Examples are shown on three almost identical checkpoints for clarity.

Keep the Sampling steps at least 25, as this is the most optimal ratio of quality and speed of generation image

As in my previous LoRa's you will need ADetailer and an additional file for it face_yolov8n.pt. For ADetailer, I usually write prompt like this: <Name_LoRa:1.0>, main trigger word(s)

Although the Negative prompt degrades the image quality, there should still be a minimum number of exceptions so that incorrect anatomy is not generated, so I created new images without the usual large Negative prompt that I use ... again, all for the sake of demonstration.