这个模型使用了超过150张高质量的原始图片进行训练。
这个模型拥有很好的画风模仿效果。
Animagine XL 和 SD1.5版本:https://civarchive.com/models/202045
illustriousXL 版本:https://civarchive.com/models/1475142
--- --- --- --- --- --- --- ---
This model used more than 150 high-quality original pictures for training.
This model has a very good imitation of the painting style.
Animagine XL and SD1.5 version:https://civarchive.com/models/202045
illustriousXL version:https://civarchive.com/models/1475142
Description
FAQ
Comments (5)
Cool!
Just 1 epoch with this performance? flux lora seems very differnet from sdxl.
@ymzlygw 确实和SDXL差别很大,训练二次元lora也很费力(不会写提示词真的很难出好图,FLUX可以完全分离人物和背景的关系,不写背景词出来的图就会比较单调,但让也就非常干净),
总结了一下:
训练某些现代风格的二次元画风就非常容易,而一些复古手绘风格的就很难;油画、水彩、水墨这些都比较容易,真人最容易
@flyx3 another thing I found is that did you train the text encoder? beacause you use some character name trigger in prompt. Now flux just support clip only for training I see.
@ymzlygw text encoder我训练了,而且使用了非常高的学习率设置,但是具体到实际情况,是否生效,其实我也说不好,因为我也还在不断尝试和调整中~













