V2 updated:
V2 utilizes a wider variety of images with different styles as training data, resulting in more diverse and enriched outcomes when generating content.
V2采用更多不同风格的图片作为训练素材，使得模型抽卡结果更加丰富！
Description
Use any checkpoint based on SD1.5
its 768,768 image
turn on Hires.fix (or not, please experiment with it yourself)
Hires steps: 14-17, denoising:04-65 ish.. just try it yourself
CGE: 5-7
Use DPM++ 2M Karras or DPM++ SDE Karras
Use 20-ish sampling steps
Strength: 0.6-1
Clip Skip: 1-3
I suggest turn off the restore faces check mark for better results of this Lora
She might look like someone from real life, but she is not what you think. She is a totally fictional character.
FAQ
Comments (1)
用一样的提示词跑不出来一样的图...