ATDAN Style LoRA
中文 | English 使用前阅读 | Read before use
--
仿ATDAN的作画风格的LoRA模型。
权重建议值:0.7~1.1
基于模型:Anything-5RE
训练信息:共计263张原画(含裁剪图),最后一次训练迭代1038步。
其他的一些建议:EasyNegative, badhandv4。

LoRA model with a painting style imitating ATDAN.
Recommended weight value: 0.7~1.1
Based on the model: Anything-5RE
Training information: A total of 263 paintings (including cropped images) were used for training, with the final training iteration completed at 1,038 steps.
Other suggestions: EasyNegative, badhandv4.
--
任何问题,欢迎反馈,问题不提出就不会得到解决。如果你创作出了好作品,欢迎分享它。如果它无法满足你的需求,我是说,我的朋友,请不要给出一星好评。
Any questions are welcome for feedback, as problems will not be solved without being raised. If you have created a great work, please feel free to share it. If it cannot meet your needs, my friend, please do not give a one-star rating.
Description
首个版本,待改进。
FAQ
Comments (3)
能问一下fitting ratio是如何评估的吗?
不好意思最近才上线,这个的评估到头来还是具有主观性的,主要用于给使用者们一个大概的参考,希望能让大家了解到我的每个模型有哪些不足和长处,判定这个模型能不能满足自己的需求。我举其中一个例子,拿肢体拟合来说,同一套参数和种子,从训练集中抽自己看得顺眼的一张图,打开它的标签文本,用这个标签使用lora和不使用lora各出一次图,看用lora跑出来的图跟训练的原图的肢体拟合得多些还是和Checkpoint当前种子的肢体拟合得多些。不断重复上述过程,计算比率得出结论。希望能够解答你的困惑。
@Chanter 感谢,我以为有自动化工具
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.








