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".

基于约300张我手裁的图片训练，因为官方角色建模还没有放出来，之后我会进一步完善，提高模型忠诚度。

提示词参考以上↑

注：

1.如果您认为模型对服装细节的还原度不够，可以添加提示词如"key necklace","keyhole on chest","leg ring","nail polish",这些是我针对大黑塔独特的服装特点，在训练时额外添加的caption。（包括钥匙形状的项链，胸口处锁孔形状的镂空，腿环，紫色指甲油）

2.大概是和底模的原因，如果用"smile"让黑塔笑起来会让人物看起来有点ooc(out of character)，所以建议用"light smile"或"light smile,closed mouth"取代"smile"。

我对LoRA训练经验不是特别丰富，如果在使用过程中发现模型有问题，欢迎批评和建议。

The Herta|大黑塔 LoRA (Detail enhanced 细节增强) - v1.1