Sign In

Two-person lora 双人lora 父女

764
5.6k
27k
31
Updated: Apr 8, 2023
animetwo-person lora
Verified:
SafeTensor
Type
LoRA
Stats
3,645
26,270
Reviews
Published
Apr 2, 2023
Base Model
SD 1.5
Trigger Words
1boy
1girl
old father
Hash
AutoV2
3E5B9D19A9
default creator card background decoration
romiolast's Avatar
romiolast
  • V3更新,解决了过拟合的问题,权重可使用0.9-1。

  • V3 update has solved the overfitting problem and the weight can now be set between 0.9-1.

/////

  • V2更新,重新训练模型,加大了稳定性,减少了多人出现的次数。

  • V2 update, retraining the model, increasing stability, and reducing the number of multiple people appearing.

原文:

这是又一个对于lora的性能测试。既上次的双人lora之后,这回lora的训练目的是希望用两组关键词同时调整两个人物。

目前看来,我达到了部分目标。男性可以用father/old father切换年龄。

实时证明,增加可控性的同时,也增加了不稳定性。我测试中大约有1/4概率会出现不想要的结果,例如:多个人、年龄差异消失。解决方案:再点一下生成。

不管怎么说,它同样能大幅度提高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0.6。其他模型和权重还未能全部测试。可以尝试结合其他的人物lora使用。

心得:1.记得用高清修复;2.其他lora适当降低权重;3.多试几次;4.负面加上(nsfw:1.2),shota,2boys,3boys,4boys

请不要尝试不合法的内容!

请在合理合法的情况下多多测试,给我反馈,谢谢!

也许我会再次优化一下这个模型。

This is another performance test for Lora. Since the last two character Lora, the goal of this Lora training session is to adjust two characters using two sets of keywords simultaneously.

So far, I have achieved some of my goals. Men can use father/old father to switch age.

Real time proof increases both controllability and instability. In my tests, there is about a quarter chance that unwanted results will occur, such as multiple individuals and age differences disappearing. Solution: Click Generate again.

In any case, it can also significantly improve the effectiveness of tag to achieve our desired goals.

I mainly use anything 4.5+weighting 0.6. Other models and weights have not yet been fully tested. You can try to use it with other characters, Lora.

Experience: 1. Remember to use Hires. fix; 2. Lower the weights of other LORA; 3. Try more times; 4. Negative plus (nsfw: 1.2), shota, 2boys, 3boys, 4Boys

Please don't try illegal content!

Please conduct more tests under reasonable and legal circumstances and give me feedback. Thank you!

Perhaps I will optimize this model again.