很高兴将我的无偏见多概念lora模型升级到flux版本。

我在civitai网站上免费下载过很多模型，因此我决定同样免费分享出我的劳动成果，来回馈大家。

（所有展示的样图都是一次性出图，没有经过高清放大或者额外优化步骤。）

PS：

我使用了接近400张无水印素材，在我可怜的4080上训练了30多个小时，来得到它。

我使用了flux.dev作为底模。

它涵盖了多人种，多肤色，多年龄，多体型，多姿态等特点。

你可以无需触发词便能很好的调用它，如果你需要一个触发词，那么它是汉语拼音：luoti。

采样步数：随便。

权重：0.5~1表现良好。

如果使用多个lora，你可以适当降低权重。

我在原始的flux.dev基础模型上进行了许多测试，在768*1024或者832*1216等我常用的分辨率下，它都表现良好。

但直到现在，我还没有测试过除了dev以外的flux其他版本，你可以自行尝试。

（缺点：由于素材过多，我使用了512*768分辨率来降低训练时间，这也导致了在某些时候，手指畸形的问题，你或许需要外接一个修复手的工作流来改善它）

===

I am very happy to upgrade my unbiased multi-concept lora model to the flux version.

I have downloaded many models for free on the civitai website, so I decided to share my work for free to give back to everyone.

(All the sample images shown are one-time output, without high-definition enlargement or additional optimization steps.)

PS:

I used nearly 400 watermark-free materials and trained more than 30 hours on my poor 4080 to get it.

I used flux.dev as the base model.

It covers multiple races, multiple skin colors, multiple ages, multiple body shapes, and multiple postures.

You can call it well without a trigger word. If you need a trigger word, it is the Chinese pinyin: luoti.

Number of sampling steps: casual.

Weight: 0.5~1 performs well.

If you use multiple loras, you can reduce the weight appropriately.

I have done many tests on the original flux.dev base model, and it performs well at my usual resolutions such as 768*1024 or 832*1216.

But until now, I have not tested other versions of flux except dev, you can try it yourself.

(Disadvantages: Due to too many materials, I used 512*768 resolution to reduce training time, which also led to the problem of finger deformity at some point. You may need to add an external hand repair workflow to improve it)