NABABA art

BY:livesting

用了两个Nababa风格的lora加几个油画类的lora做成的,其中一个已经下架了,好在之前有保存了不然也做不出这效果了,其实两个lora练得都比较过拟合,调整了好久最后只保留了上色层才搞定的,我现在已经到了瓶颈期了,自己练吧,我觉得厉害的画师,基本都很难练出好效果,等c站的lora吧有些可能练得过拟合特别严重也没法用,对人物类lora支持一般毕竟都是lora融合的了,想要两者兼得那确实是有点难为我了,可以自己尝试着用插件解决问题

就感觉如果底模一直是sd1.5后续真的就很难玩出花,毕竟来来去去都是在老旧的模型上做微调,就像一台08年的老电脑,你可以给他换机箱给他接上万的显示器,但是它内在硬件是老的,再怎么跑也跑不动大镖客

不推荐使用负嵌入和写太多负面词因为对画面影响真的很大,我的三个模型加上这个都是这样,推荐的负面提示词

(Low quality:1.2),Fuzzy,The wrong body,Draft,No details,Plane,Body error,The wrong limb,Distorted human body,Many hands and feet,No details,Chaotic structure,useless generation,Non-cue item,Children's painting,Fingers of Fusion,No torso,torso fusion,black characteristics,

It was made with two Nababa style Loras and several oil painting Loras, one of which has been removed from the shelves. Fortunately, it was saved before, otherwise this effect would not have been possible

I have reached the bottleneck period now, and I feel that if the bottom mold is always SD1.5, it will be really difficult to play with it. After all, all the adjustments are made on the old model, just like an old computer from 2008. You can replace the chassis and connect it with tens of thousands of monitors, but its internal hardware is old, and it is not competent for modern work

I do not recommend using negative embeddings or writing too many negative words because it has a significant impact on the image. My three models and this are all like this, and I recommend negative prompt words myself

(Low quality:1.2),Fuzzy,The wrong body,Draft,No details,Plane,Body error,The wrong limb,Distorted human body,Many hands and feet,No details,Chaotic structure,useless generation,Non-cue item,Children's painting,Fingers of Fusion,No torso,torso fusion,black characteristics,

我的其他模型 My Other Models

月光混合 — 夏至 - v1.0 |稳定的扩散检查点 |奇维塔伊 (civitai.com)

墨迹 - v1.0 |稳定的扩散检查点 |奇维塔伊 (civitai.com)

夏至 - 平面涂层 - v1.0 |稳定的扩散检查点 |奇维塔伊 (civitai.com)