
Recommendations:
-
With some weights it is required to set the LoRA value below 0.7, in this case it is worth specifying additional prompt parameters, for example, red hair, brown/red eye, ponytail, but for most the optimal values are from 0.7 - 1.0 and this is not required.
-
Keywords: Milly\(milly)\ or Milly\(milly\) for v3.0R, 1girl, hair ornament, brown eyes/red eyes, red hair, ponytail, hair between eyes, medium breasts.
-
CFG Scale within 5.0 - 8.0.
About versions:
What's 'D'?
Developer version
-
Just an intermediate version of LoRA.
-
Version trained on anime images.
-
Used with anything v4.5 and NED(NeverEndingDream).
-
Keywords as in release version.
What's new in version 2.0?
-
Dataset changed.
-
In LoRA v2.0R, VAE was not used during training, and other weights were used, unlike v2.0RV.
Version 1.0.
-
52 images.
-
156 steps per epoch, 14 epochs.
If you have any tips for training LoRA models, I will be glad to read.
Post your results below! ༼ つ ◕_◕ ༽つ
描述:
-
More images for training.
-
Normalized model
-
Changed the keyword to Milly\(milly\)
-
Other weights during training
-
The size of the model has become smaller
More images for training.
Normalized model
Changed the keyword to Milly\(milly\)
Other weights during training
The size of the model has become smaller
训练词语: Milly\(milly\)
名称: Milly_The_Aristocrat_s_Otherworldly_Adventure_Serving_Gods_Who_Go_Too_Far.safetensors
大小 (KB): 18551
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success