SDS_Mirror_XL

完全基于来自sd1.5模型生成图片训练的亚洲女性模型，所以会保留较明显的AI感。如果你有疑惑或者发现任何问题，可以留言也可以加入我的qq群749047075交流,密码SDS。

跑图说明：

• 起手使用1girl加黑色眼睛、嘴巴既可稳定召唤亚洲脸

• 分辨率建议 832*1152（1280）（或者其他官方训练尺寸）

• CFG:7-10 clip：2

• 采样器：DPM++2M Karras（强烈建议）在 30-40 步图片完整观感较好，往后肉眼看不出变化，最佳Restart,但略慢

• AD修脸中远景别必需，近景可以直接放大

• 推荐高清放大算法选择4x_NMKD-Superscale-SP_178000_G放大2倍，自带些许噪点，不至于放大后过于光滑，幅度0.35，步数同迭代步数。

• 手部问题可以通过叠加社区修手lora减轻。

It is entirely based on an Asian female model trained with images generated from sd1.5 models, so it retains a more obvious AI feel. If you have doubts or find any problems, you can leave a message or join my qq group 749047075 exchange, password SDS.

Running chart description:

Use 1girl with black eyes and mouth to steadily summon Asian faces

Resolution recommended 832*1152 (1280) (or other official training size)

CFG:7-10 clip: 2

Sampler: DPM++2M Karras (strongly recommended) at 30-40 steps picture complete look good, the naked eye can not see the change, the best Restart, but slightly slow

AD face repair in the long distance is necessary, close-range can be directly enlarged

It is recommended that the HD amplification algorithm choose 4x_NMKD-Superscale-SP_178000_G amplification twice, with some noise, not too smooth after amplification, the amplitude is 0.35, and the number of steps is the same as the number of iterative steps.

Hand problems can be alleviated by stacking Community hand Repair lora.