Kino-2.6d-doll style

V2版本对比V1版本没有那么容易出现糟糕的手和脚了，但是还是建议进行面部细化，因为如果面部在画幅内占据的比例过小时还是会丢失细节。V2和V1相比在细节上的处理更好一些，同时也增加对tag的理解程度，当然这并不是我的功劳，归功于我所融入的模型的训练家们。

Compared to Version 1, Version 2 is less prone to distorted hands and feet. However, facial refinement is still recommended—details tend to get lost when the face occupies a relatively small portion of the frame.

Version 2 also delivers better detail processing and improved tag interpretation compared to its predecessor. Of course, none of this is my credit; all kudos go to the creators of the models I incorporated.

没什么特别的模型，我都忘了是拿什么模型融合的了，唯一记得的是拿了don_qiaqia大佬的lora做了分层调整，反正不收buzz，喜欢的下载就好。这一版的模型质感类似于硅胶娃娃。没什么特别的提示词要求，正常的质量提示词都还不错。

提示词什么的我完全没测，不保证最终的出图效果，但是预览图都是没有加质量词和负面提示生成的，应该不会有什么低质量图出来。手脚这个完全没办法，抽卡呗还能咋办。

I haven't done any testing on the prompts at all, so I can't guarantee the final image output quality. That said, all preview images were generated without any quality-boosting keywords or negative prompts—so low-quality outputs should not be an issue. As for hands and feet? There’s really no fix for that—guess it just comes down to luck, right?

提示词什么的自己测，CFG不要太,3~~4左右，步数16~~32都还行。如果用了dmd2 Lora，最好CFG控制在0.8~~1.2，步数控制到12~~16

Test the prompts on your own. Keep the CFG scale not too high—around 3–4 works well. Step count can be set between 16–32, and the results will be decent.

If you use the dmd2 LoRA, it’s best to keep the CFG scale in the range of 0.8–1.2 and the step count between 12–16.

Recommended sampler: Euler a

Recommended scheduler: Karras

CFG:3~4

STEPS:16~32

used dmd2

CFG:0.8~1.2

STEPS:12~16

感谢：Don_qiaqia的Lora以及各位大佬的底模

Acknowledgements: Don_qiaqia's LoRA and the base models from all the amazing creators

模型类型	检查点
基础模型	Illustrious
发布时间	1/23/2026

详情

下载文件

关于此版本

模型描述

此模型生成的图像