The three previous styles in the latest version are separate styles. Because they tend to be overfitting, the actual use can reduce the weight and have a higher restoration of the original image.
Unlike the previous version, this version collects more style pictures for training, while the previous version only contains some limited pictures.
Usage Notes: Based on noobaiXLNAIXL_epsilonPred11Version.safetensors training, please use it based on this version.
Weight: 1,
Trigger words are tags from dan, you need to add, some styles are still included even if lora is not used, noob still contains it, please add it as needed, among which konya_karasue, is very inconspicuous, so it is very good.
The main style words included in the training are as follows:
konya_karasue,
qtian,
torino_aqua,
zoirun,
dino_ (dinoartforame),
iumu,
da_mao_banlangen,
shibuki_kamone,
and some chibi styles.
Description
trained base on noobaiXLNAIXL_epsilonPred11
example run on noobaiXLNAIXL_epsilonPred11
trigger word：artist name,
style2：qtian,
weight：1,<=1,
其实不是很达预期，且数据集的风格并不一致，不相像。