
,这是我练的一个LORA,采集来源KS,训练图片23张,尺寸768*768,图片数量*10*30/*2
,该模型泛用性极高,适用各种大模型
,DPM++ 2M Karras 20-28
,建议0.5~0.8
,0.5~0.8建议开启ADetailer
,0.5泛化性高加上修复脸部给到0.8权重+ADetailer修复
,高清采用4x_NMKD-Siax_200k,10步,0.25-0.4
,本人炼制真人模型有一定的经验可以还原素材80%-95%
,接受真人定制(本人只练真人,其他一概不懂,,,)
,如需要数据集/训练参数,请联系本网站私聊
,更多福利可以移步CCCCC,或联系本人
,发布的LORA请多多返图,感谢支出
,出品不易,欢迎讨论,学习,交流
,QQ:39826676(福利)
. This is a LORA I practiced, collected from KS, 23 training pictures, size 768*768, Number of pictures
1030/*2
This model is extremely versatile and suitable for various large models
. DPM++ 2M Karas 20-28
. Recommend 0.5~0.8
. 0.5~0.8 It is recommended to open ADetailer
0.5 High generalization plus repair and repair gives a weight of 0.8 + ADetailer repair
. HD uses 4x_NMKD-Siax_200k, 10 steps, 0.25-0.4
I have some experience in refining real-life models and can restore 80%-95% of the original materials.
. Accept customization by real people (I only practice real people and know nothing else...)
. If you need data set/training parameters, please contact this website for private chat
For more benefits, please go to CCCCC, or contact me
. Please return more pictures after publishing LORA, thank you for your expenditure
. It’s not easy to produce, welcome to discuss, learn and communicate
.QQ:39826676(Welfare)
描述:
。这是我练的一个LORA,采集来源KS,训练图片23张,尺寸768*768,图片数量*10*30/*2
。该模型泛用性极高,适用各种大模型
。DPM++ 2M Karras 20-28
。建议0.5~0.8
。0.5~0.8建议开启ADetailer
。0.5泛化性高加上修复脸部给到0.8权重+ADetailer修复
。高清采用4x_NMKD-Siax_200k,10步,0.25-0.4
。本人炼制真人模型有一定的经验可以还原素材80%-95%
。接受真人定制(本人只练真人,其他一概不懂。。。)
。如需要数据集/训练参数,请联系本网站私聊
。更多福利可以移步CCCCC,或联系本人
。发布的LORA请多多返图,感谢支出
。出品不易,欢迎讨论,学习,交流
。QQ:39826676(福利)
. This is a LORA I practiced, collected from KS, 23 training pictures, size 768*768, Number of pictures
1030/*2
This model is extremely versatile and suitable for various large models
. DPM++ 2M Karas 20-28
. Recommend 0.5~0.8
. 0.5~0.8 It is recommended to open ADetailer
0.5 High generalization plus repair and repair gives a weight of 0.8 + ADetailer repair
. HD uses 4x_NMKD-Siax_200k, 10 steps, 0.25-0.4
I have some experience in refining real-life models and can restore 80%-95% of the original materials.
. Accept customization by real people (I only practice real people and know nothing else...)
. If you need data set/training parameters, please contact this website for private chat
For more benefits, please go to CCCCC, or contact me
. Please return more pictures after publishing LORA, thank you for your expenditure
. It’s not easy to produce, welcome to discuss, learn and communicate
.QQ:39826676(Welfare)
训练词语: A 24_year_old_girl_named_Zaizai
名称: 仔仔2-014.safetensors
大小 (KB): 147571
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success