Type | |
Stats | 249 57 |
Reviews | (36) |
Published | Nov 7, 2023 |
Base Model | |
Training | Steps: 1,800 Epochs: 30 |
Usage Tips | Clip Skip: 7 Strength: 0.8 |
Trigger Words | maestrale \(azur lane\), multicolored hair, white_hair, red_hair, bangs, red_eyes, very_long_hair, open_jacket, beret, cropped_jacket, green_jacket, green_headwear, long_sleeves, hair_bow, stripped_bow, strapless dress, white_dress, knee boots, black_thighhighs, detached_collar, armband red_ahoge, two-tone hair, bangs_pinned_back, forehead |
Hash | AutoV2 C215F9D470 |
这是西北风，一个碧蓝航线中的角色，原型为意大利海军西北风级驱逐舰1号舰
与她的妹妹西南风类似，这个LoRA的关键词为maestrale \(azur lane\), multicolored hair, white_hair, red_hair, bangs, red_eyes, very_long_hair, open_jacket, beret, cropped_jacket, green_jacket, green_headwear, long_sleeves, hair_bow, stripped_bow, strapless dress, white_dress, knee boots, black_thighhighs, detached_collar, armband
由于她的发型比较特殊，所以你通常需要通过提高multicolored hair的权重，或者添加例如red_ahoge, two-tone hair, bangs_pinned_back, forehead等提示词来更好地还原她的发型
值得一提的是，全身图会使得她的发型特点更难以很好地生成，而相对应的，半身图的表现则会更好，之前的西南风也是如此，所以请权衡利弊
当然了，记得根据实际情况灵活调整提示词
本LoRA以AnyLoRA为底模进行训练
推荐权重为0.8左右，建议不超过1.0
极度推荐使用LoRA Block Weight进行分层，推荐预设为1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,0,0，当然，如果你有更好的预设，欢迎分享
和她的妹妹一样平
但是谁能拒绝白毛红瞳傲娇小萝莉驱逐舰妹妹呢
This is Libeccio, a character in Azur Lane, based on the Italian Navy's Libeccio-class destroyer lead ship.
Similar to her sister Maestrale, the key words for this LoRA are maestrale (azur lane), multicolored hair, white_hair, red_hair, bangs, red_eyes, very_long_hair, open_jacket, beret, cropped_jacket, green_jacket, green_headwear, long_sleeves, hair_bow, stripped_bow, strapless dress, white_dress, knee boots, black_thighhighs, detached_collar, armband.
Due to her unique hairstyle, you may need to increase the weight of "multicolored hair" or add additional keywords such as "red_ahoge," "two-tone hair," "bangs_pinned_back," "forehead" to better recreate her hairstyle.
It's worth mentioning that generating a full-body image may make it more difficult to accurately depict her hairstyle, while a half-body image may work better. This was also the case with Maestrale, so please consider the pros and cons.
Remember to adjust the keywords flexibly based on the actual situation.
This LoRA is trained based on AnyLoRA.
The recommended weight is around 0.8, but it is not recommended to exceed 1.0.
It is highly recommended to use LoRA Block Weight for hierarchical training, with the recommended preset being 1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,0,0. Of course, if you have a better preset, feel free to share.