Two-person-lora-lora-Update-for-character-Lora

我要开发同款
匿名用户2024年07月31日
11阅读
开发技术Pytorch
所属分类ai、text-to-image、lora
开源地址https://modelscope.cn/models/sd_lora/Two-person-lora-lora-Update-for-character-Lora
授权协议Apache License 2.0

作品详情

About v2:

做了优化,解决了过拟合的问题,现在可以不下调其他lora的权重,更好地适配角色lora了!寻找属于你的姐姐吧!

感谢hako-mikan!

感感谢大江户战士。

心得:使用tall girl,big breast等维持姐姐。

看在姐姐的份上,给我个5星如何:D

Optimization complete! Overfitting resolved, no need to adjust other LORA weights. Now better suited for character Lora. Find your perfect big sister!

Thanks to hako-mikan,

Thanks to CyberAIchemist

Tips: Using traits like 'tall girl' and 'big breasts' to maintain the 'big sister' theme.

How about giving me a 5-star rating as a favor to your sister? :D

原文:

这是一个对于lora的性能测试 。也许是第一个针对双人的lora?(如果有人见过请告诉我)。这回lora的训练目的是不针对外貌或者动作,而是针对两个角色的关系。所以我选择了这样一个关系(看trigger words)。

目前看来,效果是有的,但是也并不绝对稳定,少部分时候会出现黑白、单人等情况。不过它已经能大幅度提高tag的效果,以达到我们想要的目的。

我主要使用的是anything4.5+权重0.7。其他模型和权重还未能全部测试。可以结合其他的人物lora使用。

心得:1.记得用高清修复;2.其他lora适当降低权重;

请多多测试,给我反馈,谢谢!

This is a performance test for Lora. Perhaps it's the first test for a Two-person Lora? (Please let me know if anyone has seen one before.) The training purpose of this Lora is not focused on appearance or movements, but on the relationship between two characters. So I chose this kind of relationship (see trigger words).

For now, the results are good, but not completely stable. In some cases, it may show in black and white or only one person. However, it has greatly improved the effectiveness of the tags, achieving the goal we wanted.

I mainly use anything4.5 with weight 0.7. I haven't been able to test all the other models and weights yet. You can combine it with other character Lora.

Tips: 1. Remember to use Hires. fix ; 2. Lower the weights of other LORA.

Please test it more and give me feedback. Thank you!

声明:本文仅代表作者观点,不代表本站立场。如果侵犯到您的合法权益,请联系我们删除侵权资源!如果遇到资源链接失效,请您通过评论或工单的方式通知管理员。未经允许,不得转载,本站所有资源文章禁止商业使用运营!
下载安装【程序员客栈】APP
实时对接需求、及时收发消息、丰富的开放项目需求、随时随地查看项目状态

评论