NewChinaStyle版本v1.0 (ID: 195318)

这个lora可以彻底的改变大模型的画风,

1、优化了tag，能更好的文生图；

• Optimized tags for better text-to-image generation.

2、优化了数据集，生成的图像更精美，更精致；

• Optimized dataset for more beautiful and refined images.

3、对部分叠加模型支持的不好，推荐使用原生模型，如：万象熔炉 | Anything V5/V3

• Not well supported for some overlay models, recommended to use native models, such as Wanxiang Fuwu | Anything V5/V3.

4、开启高清修复可以让画面更丰富，关闭高清修复可以让画面更简洁；

• Turning on high-resolution restoration can make the image richer, turning off high-resolution restoration can make the image simpler.

5、高清修复时，如果你用其他放大算法得不到好的效果，可以试试SwinlR 4x算法；

• When using high-resolution restoration, if you don't get good results with other upscaling algorithms, you can try SwinlR 4x algorithm.

6、提示词要精准，不要输入一些你自己都想不出能生成什么画面的提示词，主画面尽量用自然语言输入，如：一只戴着太阳镜的招财猫，手里拿着棒棒糖；

• The prompt words should be accurate, don't enter prompt words that you can't think of what kind of image to generate, the main screen should be input in natural language as much as possible, such as a lucky cat wearing sunglasses, holding a lollipop in hand.

7、lora的权限根据大模型的不同，可以从0.7~1.1，权限越低画面受到lora的影响越小，但是有时候效果会变好；

• The authority of lora can be from 0.7 to 1.1 according to different large models, the lower the authority, the less the image is affected by lora, but sometimes the effect will be better.

8、内置了VAE，出图时如果效果不好，可以关闭VAE；

• VAE is built in, if the output image is not good, you can turn off VAE.

描述:

这个lora可以彻底的改变大模型的画风。

1、优化了tag，能更好的文生图；

• Optimized tags for better text-to-image generation.

2、优化了数据集，生成的图像更精美，更精致；

• Optimized dataset for more beautiful and refined images.

3、对部分叠加模型支持的不好，推荐使用原生模型，如：万象熔炉 | Anything V5/V3

• Not well supported for some overlay models, recommended to use native models, such as Wanxiang Fuwu | Anything V5/V3.

4、开启高清修复可以让画面更丰富，关闭高清修复可以让画面更简洁；

• Turning on high-resolution restoration can make the image richer, turning off high-resolution restoration can make the image simpler.

5、高清修复时，如果你用其他放大算法得不到好的效果，可以试试SwinlR 4x算法；

• When using high-resolution restoration, if you don't get good results with other upscaling algorithms, you can try SwinlR 4x algorithm.

6、提示词要精准，不要输入一些你自己都想不出能生成什么画面的提示词，主画面尽量用自然语言输入，如：一只戴着太阳镜的招财猫，手里拿着棒棒糖；

• The prompt words should be accurate, don't enter prompt words that you can't think of what kind of image to generate, the main screen should be input in natural language as much as possible, such as a lucky cat wearing sunglasses, holding a lollipop in hand.

7、lora的权限根据大模型的不同，可以从0.7~1.1，权限越低画面受到lora的影响越小，但是有时候效果会变好；

• The authority of lora can be from 0.7 to 1.1 according to different large models, the lower the authority, the less the image is affected by lora, but sometimes the effect will be better.

8、内置了VAE，出图时如果效果不好，可以关闭VAE；

• VAE is built in, if the output image is not good, you can turn off VAE.

训练词语: xgc