有事加群：697080799,188546326

示例图我尽力了，它的实际效果要你们自己使用。

I tried my best with the example picture, but you have to use it for your own actual effect.

作为最初版本，它对其他lora的兼容非常不好，但是图片更真实，对不同年龄的反应也更灵敏

As the initial version, its compatibility with other loras is very poor, but the pictures are more realistic and more responsive to different ages.

————————————————————————————————————————

Training objectives:

1. Minimize the greasy or internet celebrity-like feel in Asian models.

2. Enable controllable age adjustments to enhance the realistic body proportions of anime characters.

3. Ensure compatibility with various photographic styles.

4. Do not lock facial features.

训练目的:

1.尽量去掉亚洲模型的油腻、网红感。

2.可控年龄来提高对二次元人物写实化的身材比例问题。

3.兼容各种摄影风格。

4.不锁脸。

I trained a model called "kolo-realistic" in a bug. Currently, it is exclusive to the "lib," and it will be released to "civit" in a week later. What's being updated here now is "kolo-realistic-ani." Compared to the former, "ani" has increased compatibility with two-dimensional content, but it sacrifices some degree of realism.

https://www.liblib.ai/modelinfo/0553cdb40cb64293b62226ad6e76c9ad

我在bug中训练了一个模型kolo-realistic，目前在lib独家，一周后会发到civit，现在这里更新的是kolo-realistic-ani，ani与前者相比，增加了对二次元的兼容性，但是牺牲了一定的写实。

I would be happy if you could add it to "lib" for download. It would also motivate me to further optimize it, enabling it to draw men, girls of different age groups (Asian), and adapt to a wider range of lighting and artistic styles. If not many people use it, I might become less motivated to update it. After all, right now, with just a few thousand images in the dataset, it takes me dozens of hours. I hope to expand the dataset to tens of thousands of images to cater to a broader audience.

如果你们能去lib增加他的运行我会很高兴，也有更多的动力去优化它，让它可以画男人，画各个年龄段的女孩子，（亚洲人）适应更多的光影风格。如果没什么人使用，我大概就懒得更新了，毕竟现在只要几千张数据集我就需要几十个小时，我希望将数据集提高到几万张来扩展到不同的人群。

Of course, it's still part of the "kolo" series. Friends who have downloaded previous versions should understand what I mean. You can control the character's age using tags like 5yo, 10yo, 20yo, 30yo (for specific usage instructions, check the section on "kolo-yma" below). This allows for better control over the realistic body proportions of animated characters. Unfortunately, Civit won't allow me to release images of characters like nahida and klee.

当然它依然是kolo系列，下载过前面版本的朋友应该懂我的意思，通过5yo,10yo,20yo,30yo这些tag去控制人物的年龄，（具体使用方法去看下面kolo-yma的部分）这使得它对动漫人物写实化的身材比例有了更好的控制。很可惜，civit不会让我发出纳西妲，可莉的图片。

Using "masterpiece, best quality," will result in a closer approximation to everyday smartphone photos,

while using "masterpiece, best quality,8K,realistic, High definition, photography, depth of field,HDR, highres,photoshop \(medium\)" will achieve a more photographic effect.

You can increase the weight of "photoshop \(medium\)" as needed to enhance this effect.

使用“masterpiece, best quality”会更接近日常手机拍照，

而使用“masterpiece, best quality,8K,realistic, High definition, photography, depth of field,HDR, highres,photoshop \(medium\)”会更接近写真效果，

你可以适当提高“photoshop \(medium\)”的权重来增加这种效果。

Below is my rambling. In the end, this thing is meant for training my anime character Lora in a realistic style. It has been featured on the front page of "lib" for a day, but there hasn't been much usage. It could be that I'm really not doing a good job with the training, or perhaps this style isn't popular. If that's the case,I might as well go back to training 'Touhou' characters and retire from this.

下面是我的碎碎念，说到底，这玩意是为了我训练的动漫角色lora跑写实风格用的，在lib上首页推荐了1天也没啥人使用，可能是我真的训练的不好，或者这个风格没人喜欢。如果结果如此，我也只能回去继续练touhou角色养老了。

————————————————————————————————————————

kolo-yma

1. Yes, yma has retained the core components of kolo-v1, but now it can produce faces of different ages.

2. Using "5yo, ch***," "10yo, lit*** girl," "20yo, mature female" can yield better results. In any case, I recommend including "yo" for control. At least in my experience with the realistic model, it can achieve the effect of generating a face every 2 years.

3. However, this also leads to yma not performing as well when generating v1-style images. I'm not sure if this direction is correct. Please tell me, would you prefer optimizations for the yma version or the v1 version?

4. The current compatibility with LoRa is not very satisfactory, and I won't address this issue until I have created a version that satisfies the stylistic requirements.

1.是的，yma保留了kolo-v1的核心组件，但是现在可以出现不同年龄

2.使用"5yo,ch***","10yo,lit*** girl","20yo,mature female"可以得到更好的效果，总之建议你加入"yo"来进行控制，至少我在写实模型中，它可以做到每2yo一张脸的效果。

3.但是，这也导致在使用yma版本跑v1风格图片时，效果不是那么好了，我不知道这个方向是否正确，请告诉我，你更想要yma版本的优化，还是v1版本的优化。

4.目前它对于lora的兼容效果不是很好，直到做出风格上满意的版本为止，我暂时不会去解决这个问题。

————————————————————————————————————————

kolo-v1

建议使用负面
Negative prompt
(blurry,blurry background:1.4),nsfw,nude,text,username,logo,(low quality, worst quality:1.4), (bad anatomy), (inaccurate limb:1.2), bad composition, inaccurate eyes, extra digit, fewer digits, (extra arms:1.2),

1.本模型训练的初衷，是为了生成七七，可莉，纳西妲这类角色时，可以有更真实的身材比例。

2.本模型制作过程中未使用任何nsfw图片。

3.为了防止被滥用，我努力将其封印在了二次元画风中。所以会导致一些问题，比如有时会很模糊，有时头发的颜色需要你提高权重。而且他也画不了男人了。

4.由于加入了些功能性lora，比如皮肤质感，导致有时人物衣服会较少，需要你使用tag提示。

5.请勿使用此模型生成nsfw相关图片！后果自负！

6.如果civit没有删除这个模型，后续会更新多个版本，实际上在制作中有多个版本效果更好，但是为了防止滥用，暂时选取此版本。

1. The original intention of training this model is to have more realistic body proportions when generating characters such as Qiqi, Klee, and Nahida.

2. No nsfw pictures were used in the production of this model.

3. In order to prevent abuse, I tried hard to seal it in a two-dimensional painting style. So that causes some problems, like sometimes it's very blurry, and sometimes the hair color requires you to increase the weight.And he can no longer draw men.

4. Due to the addition of some functional loras, such as skin texture, sometimes the characters will have less clothes, and you need to use tag prompts.

5. Please do not use this model to generate nsfw related pictures! Do it at your own risk!

6. If civit does not delete this model, multiple versions will be updated in the future. In fact, there are multiple versions in production that are better, but in order to prevent abuse, this version is temporarily selected.