LoRA属于微调模型,但是有很强烈的风格化倾向,因此有LoRA和没有LoRA,差别非常大。下面介绍几个最为流行的LoRA。
LoRA的使用方法有直接使用和插件使用两种,前面的系列文章中已经介绍过,不再赘述。
常见LoRA推荐和介绍
墨心 MoXin & 疏可走马
墨心是由安吉吴仓石、兴化板桥先生、八大山人、山阴伯年等大师之大小写意作品辅以现代人物训练而成。辅以恰当之提示词,诵先贤尊号,襄古今并用之意,明雅俗共举之美。
疏可走马是一个和墨心搭配使用的构图Lora, 一旦使用并再最前前置提示词后,就会采用较大面积留白的构图风格。
注意事项:
1.)CFG范围将会改变风格
1~3 : 大小写意
3~7 : 逐渐工笔
2.)推荐基础模型为ChilloutMix、国风3.2等
3.)《墨心》的推荐Lora权重为0.85以下
4.)《疏可走马》推荐Lora权重为0.7~1
模型:Chilloutmix
LORA: 墨心 MoXin 权重0.8 & 疏可走马 权重0.8
提示词:shukezouma, negative space, , shuimobysim , <lora:shuV2:0.8>, portrait of a woman standing , willow branches, (masterpiece, best quality:1.2), traditional chinese ink painting, <lora:shuimobysimV3:0.7>, modelshoot style, peaceful, (smile), looking at viewer, wearing long hanfu, hanfu, song, willow tree in background, wuchangshuo,
反向提示词:(worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, skin spots, acnes, skin blemishes, age spot, glans, (watermark:2),
blindbox/大概是盲盒
盲盒风格的LORA,非常可爱,风格化也特别明显。
关键词:full body,chibi,
权重建议:1
大模型: RevAnimated 或者其他你喜欢的都行(不同模型出来的结果差异较大,建议使用写实风格类的模型,但是用其他模型也会有不一样的画风出来,也值得一试)
模型:deliberate_v2
模型:anything-v4.0-pruned
提示词:
(masterpiece),(best quality),(ultra-detailed), (full body:1.2),
1girl,chibi,cute, smile, open mouth,flower, outdoors, playing guitar, music, beret, holding guitar, jacket, blush, tree, :3, shirt, short hair, cherry blossoms, green headwear, blurry, brown hair, blush stickers, long sleeves, bangs, headphones, black hair, pink flower,(beautiful detailed face), (beautiful detailed eyes), <lora:blindbox_v1_mix:1>,
反向提示词: (low quality:1.3), (worst quality:1.3)
KoreanDollLikeness/Taiwan doll likeness
配合Chilloutmix使用的亚洲美女LORA,注意法律风险,请勿用于非法用途。
模型:chilloutmix
提示词:best quality, ultra high res, (photorealistic:1.4), 1girl, off-shoulder white shirt, black tight skirt, black choker, (faded ash gray hair:1), looking at viewer, closeup
反向提示词:paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), skin spots, acnes, skin blemishes, age spot, glans,
Anime Lineart / Manga-like (线稿/線画/マンガ風/漫画风) Style
漫画风、线稿风格的LORA,特点也很强。
推荐配置:
大模型:Anything 系列
VAE:Orangemix(同NAI)
LoRA强度:1
采样器:DPM++ 2M Karras
采样步数:20
CFG:7
负面embedding:EasyNegative、badhandv4
推荐使用高清修复。
模型:Anything 4.0
提示词:masterpiece, best quality, 1girl, solo, long_hair, looking_at_viewer, smile, bangs, skirt, shirt, long_sleeves, hat, dress, bow, holding, closed_mouth, flower, frills, hair_flower, petals, bouquet, holding_flower, center_frills, bonnet, holding_bouquet, flower field, flower field, lineart, monochrome, <lora:animeoutlineV4_16:1>
反向提示词:EasyNegative, badhandv4
Miniature world style 微缩世界风格
这是一个LyCORIS模型,也可以归为Lora模型,也是属于微调模型的一种。使用方法详见前文关于LyCORIS模型的介绍。
模型:revAnimated_v11
提示词:mini\\(ttp\\), (8k, RAW photo, best quality, masterpiece:1.2), cyberpunk city, miniature, landscape, isometric, <lora:miniature_V1:0.8>
反向提示词:(((blur))), (EasyNegative:1.2), ng_deepnegative_v1_75t, paintings, sketches, (worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality, ((monochrome)), ((grayscale)), bad anatomy,(long hair:1.4),DeepNegative,(fat:1.2),facing away, looking away,tilted head,lowres,bad anatomy,bad hands, text, error, missing fingers,extra digit, fewer digits, cropped, worstquality, low quality, normal quality,jpegartifacts,signature, watermark, username,blurry,bad feet,cropped,poorly drawn hands,poorly drawn face,mutation,deformed,worst quality,low quality,normal quality,jpeg artifacts,signature,watermark,extra fingers,fewer digits,extra limbs,extra arms,extra legs,malformed limbs,fused fingers,too many fingers,long neck,cross-eyed,mutated hands,polar lowres,bad body,bad proportions,gross proportions,text,error,missing fingers,missing arms,missing legs,extra digi
LoRA模型使用注意事项
1. LoRA模型很容易训练某些特定的人物、形象或者事物,请切记不要损坏原有版权所有人的权益,尤其是要注意模型是否可以商用。如果不清楚,建议向版权所有者咨询。
2. 模型禁止用于训练基于明星、公众人物肖像的风格模型训练,因为这会带来争议,对AI社区的发展造成不良的负面影响。
3. 有些模型争议较大,因此需要大家根据自身实际需求和情况自行下载(记得科学上网哦~),但是切记不要用于网络诈骗、传播黄色淫秽作品等非法目的。SD是开源软件,给了大家很大的自由度,但是切记网络并非法外之地。有些作品自己欣赏不犯法,但是传播就犯法了,各位谨记。