More in-depth optimization has been carried out for clothing details in the latest version.
Trained on 300 self-cropped and removed bg images.Since I'm not very experienced in LoRA training,if you find problems with the model during use, please let me know your criticism and advice.
weight: about 0.95
prompt:
(best quality,8K, masterpiece ,simple flat background, clear background,none background),mihoyo,1girl, solo, hat, long hair, detached sleeves, jewelry, dress, witch hat, earrings, flower, bangs, hat flower, necklace,key,bare shoulders, black dress, black footwear,frilled sleeves, black headwear, grey hair, frills, long sleeves,cross, purple flower,stockings,boots,knee boots
Notes:
If you find that the LoRA does not accurately reproduce the clothing details, consider adding prompts like "key necklace," "keyhole on chest," , "leg ring","nail polish". These words were additionally included in the captions of the dataset images during training.
（Including key-shaped necklace, keyhole-shaped hollow on the chest, leg rings, purple nail polish）
Due to the base model, using word "smile" to make Herta smile may result in OOC (out of character). Therefore, it is recommended to use "light smile" or "light smile, closed mouth" instead of "smile".