V3更新：
恭喜绮良良正式上线了！草猫猫可爱捏~
V3版本扩充了数据集。
加强了换装和NSFW的能力。
加强了绘制猫足的能力，虽然还是不太好，请反复尝试！
使用方法：
推荐权重：0.8~0.6
触发词：
1girl, green eyes
official, two tails, paw （原版服装）
translate by GPT：
V3 Update:
Congratulations to Kirara for officially launching on the V3 version! The grass cat is so cute~
The V3 version has expanded its dataset,
Enhance the abilities of clothing changing and NSFW.
The ability to draw cat paws has also been improved, although still not perfect. Please try repeatedly!
Usage:
Recommended weight: 0.8~0.6
Trigger words:
1girl, green eyes
official, two tails, paw (original costume)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
V2更新：
今天有实机3d演示，以此扩充了数据集（也加入了一些nsfw的部分），重新进行了训练，降低了过拟合的情况，姿势更加多样，换装仍然存在一定困难
使用方法：
推荐权重：0.7~0.9
推荐触发词：1girl, official, green eyes, ahoge, hair ornament, bare shoulders, black crop top, navel, detached sleeves, elbow gloves, fingerless gloves, tail, skirt
translate by GPT：
Version 2 Update:
Today, there was a real-time 3D demonstration to expand the data set (including some NSFW content), and the model was retrained to reduce overfitting. The poses are more diverse, but there are still some difficulties with changing outfits.
Usage:
Recommended weights are between 0.7 and 0.9,
Recommended trigger words：1girl, official, green eyes, ahoge, hair ornament, bare shoulders, black crop top, navel, detached sleeves, elbow gloves, fingerless gloves, tail, and skirt
Description
V2更新：
今天有实机3d演示，以此扩充了数据集（也加入了一些nsfw的部分），重新进行了训练，降低了过拟合的情况，姿势更加多样，换装仍然存在一定困难
使用方法：
推荐权重：0.7~0.9
推荐触发词：1girl, official, green eyes, ahoge, hair ornament, bare shoulders, black crop top, navel, detached sleeves, elbow gloves, fingerless gloves, tail, skirt
translate by GPT：
Version 2 Update:
Today, there was a real-time 3D demonstration to expand the data set (including some NSFW content), and the model was retrained to reduce overfitting. The poses are more diverse, but there are still some difficulties with changing outfits.
Usage:
Recommended weights are between 0.7 and 0.9,
Recommended trigger words：1girl, official, green eyes, ahoge, hair ornament, bare shoulders, black crop top, navel, detached sleeves, elbow gloves, fingerless gloves, tail, and skirt
FAQ
Comments (6)
使用相应LORA的结果
确认了维持格子point模特的画风 塑造了角色 我非常感叹
3d模型利用mmd使用的模型进行了train吗?
想知道以什么方式使用了3d图像。
您是否使用 crop 来优化图像？
瞳孔细节不结块，非常美丽。
如果是失礼的问题，我会删除的
Hello, due to some translation issues, I may not fully understand your message, but I will try my best to answer and hope to be helpful to you.
Firstly, regarding the use of 3D models, screenshots of 3D models account for approximately 20-30% of the dataset. They are very helpful in filling in different angles of characters, such as back and side views. I also repeated partial screenshots to complete details.
Secondly, I did not use the cropping function but enabled bucket, which automatically groups images of different sizes.
It is worth noting that I once cropped the dataset to a fixed size and removed the background, but the result was not good, and the model tended to learn the style of cropping and background removal.
If you have any other questions, please feel free to ask in English.
@Pumpkinhn Hello 大哥
大哥's latest Star Rail LORA, it is possible to implement a high level
To train using 3D tools(Koikatu?)
I am curious about the running 3d tool and 3d model pose and how to create the dataset.
@yh12 Hello,
You can take screenshots within the game instead of using 3D modeling software.
Capture different angles of the model, such as front, side, and back, and include close-up shots of specific parts like the head and clothing.
The key is to have a comprehensive range of angles, with attention to detail and avoid having an excessively high proportion of the total dataset to avoid a consistent art style.
If the training results are unsatisfactory, consider learning about pre-processing tag data and adjusting parameters.
Do you mind sharing a link to the model you used for the preview images? it says "二次元_CounterfeitV25_25" I am guessing that is different from Counterfeit-V2.5?
this please <3
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.