镭射衣模型，触发词（leishe,laser）
Laser suit model, trigger word (leishe,laser)
have fun! :D
Version230626:之前有朋友反馈说生成的人物头在画面外，第一个版本训练素材好多都是只有衣服，导致ai学习了素材的构图。这个版本用512*768训练的，构图方面有所改善。不足之处还请多指出，也希望大佬们多多指导。祝大家玩得开心~😉
Previously, some friends provided feedback that the generated characters' heads were outside the frame. The first version of the training data mostly consisted of clothing only, which resulted in the AI learning the composition of the frame. This version was trained with a resolution of 512*768, and there have been improvements in composition. Please point out any shortcomings and provide more guidance. I also hope that experts can provide more guidance. I wish everyone a fun time!（Translated by chatgpt）😎
PS:这个版本大部分训练素材和第一个版本都不一样，生成效果也不太一样，如果觉得不满意可以结合着第一个版本分配权重一起用。The training data for this version is mostly different from the first version, and the generated results are also different. If you're not satisfied, you can combine it with the first version and adjust the weights accordingly.
Description
之前有朋友反馈说生成的人物头在画面外，第一个版本训练素材好多都是只有衣服，导致ai学习了素材的构图。这个版本用512*768训练的，构图方面有所改善。不足之处还请多指出，也希望大佬们多多指导。祝大家玩得开心~😉
Previously, some friends provided feedback that the generated characters' heads were outside the frame. The first version of the training data mostly consisted of clothing only, which resulted in the AI learning the composition of the frame. This version was trained with a resolution of 512*768, and there have been improvements in composition. Please point out any shortcomings and provide more guidance. I also hope that experts can provide more guidance. I wish everyone a fun time!（Translated by chatgpt）😎
PS:这个版本大部分训练素材和第一个版本都不一样，生成效果也不太一样，如果觉得不满意可以结合着第一个版本分配权重一起用。The training data for this version is mostly different from the first version, and the generated results are also different. If you're not satisfied, you can combine it with the first version and adjust the weights accordingly.
FAQ
Comments (6)
I'm really glad about the upgrade. The issue I mentioned seems to be solved. But...got another ti~~~ny issues. 1.The Character that i generated has a pot belly. In front view. Not always, but pretty stand out....2.When lora weight high (about over 0.8), the quality of the image is degraded (pixelated). Please add description recommend weigt, setting and how to use for newbie(like me). Could you check out? And last, If you have time...I think i could adjust the body shape,clothes with this lora. Mybe this causes is my lack of stableDiffusion skill. Anyway thank's for sharing.
这个lora模型与各个基础模型的适配性有些差距，需要测试并调整权重来找到一个比较好的效果。我也是个业余玩家，还在不断学习新技能，有些问题可能需要再去不断完善。谢谢您的反馈与支持。
There are some gaps in the suitability of this lora model to each base model, and it needs to be tested and the weights adjusted to find a better result. I am also an amateur player, still learning new skills, some problems may need to go back to constantly improve. Thank you for your feedback and support (机翻Translate with deepl)
大佬，修改后的训练方法是加了带头的训练素材吗？这样会不会影响人脸lora？
是的，确实用了带头素材。应该多多少少会有些影响吧，我还真不知道咋能排除训练素材中脸的影响。等学到新方法了再更新吧。
@Someone97421 把素材图的脸直接p掉或者p成表情那种简笔画会不会有用
@klhscott 哈哈这个思路也可以试试，本来想有时间了研究研究lora分层控制融合啥的。
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.