Type | |
Stats | 171 37 |
Reviews | (16) |
Published | Oct 7, 2024 |
Base Model | |
Training | Steps: 2,000 |
Usage Tips | Strength: 0.9 |
Trigger Words | Furina_Genshin |
Hash | AutoV2 993F92AB5E |
芙宁娜·德·枫丹(仅动漫风),提供4000-steps和2000-steps两个版本
Furina de Fountaine (Anime Only Lora), providing two versions of 4000-steps and 2000-steps
4000-steps 能够提供更贴合设计的细节,但是由于过拟合问题,有时画面结构和肢体细节会崩。
4000-steps can provide more design-friendly details, but due to overfitting, the picture structure and body details are sometimes corrupted.
2000-steps 差异并不明显,但是能够适配更多画风、动作,同时画面结构和肢体细节更加稳定。
2000-steps has no obvious difference, but can adapt to more painting styles and actions, and the picture structure and body details are more stable.
由于训练数据集混入了部分芙宁娜和黄毛的双人作品(笑),偶尔会产生黄发芙宁娜。
Since the training data set contains some duets of Funina and somewhat yellow-haired guy (LoL), yellow-haired Funina will occasionally appear.
对异色瞳的支持还在研究中,暂且命名为version 0.1,现在看来在 prompt 中加入 Heterochromia eyes 有概率产生异色瞳。
Support for heterochromia eyes is still under research, temporarily named version 0.1. Now it seems that adding "Heterochromia eyes" to the prompt may result in heterochromia eyes.
对 Cosplay 的支持还在收集数据,我不希望一个cosplay模型训练出来最后都是同一个coser的脸,最好能够生成亚洲和法国两种面部特征的变体。
Still collecting data for Cosplay support. I don’t want a cosplay model to be trained with the same cosplayer’s face in the end. It would be best if it could generate variants of both Asian and French facial features.