谢谢！使用后喜欢的话打个分哦。

Thanks

Of course.

Hi. I had seen your nice picture. You can try (NegfeetV2, NegfeetV2:1.2) instead of (NegfeetV2:1.6), this may perform similar without loss image quality. On the other hand, if you find you must use high weight to generate nice feet, the checkpoint you're using may not be good for feet by itself, as well as when generating soles, few checkpoints performs good enough, that is the issue I'm keeping on solving. Try higher sample steps or using DDIM sampler, even, you can try lower Negfeet weight during highres fix.

@Svensson As my understanding, it means to double the token lenth of NegfeetV2, then make the doubled embedding weight 1.2.

谢谢！

你好，我刚进行了白丝的测试，完全是没有问题的,你可以在评论区的图片里看到我的测试。请问你使用的是什么checkpoint？是否将这个embedding与其它负面embedding进行了混合使用？

@bilibilissr 你好，我测试了一下黑丝，的确发现了你说的脚踝处可能出现皮肤的情况。这个问题可以通过prompt比较容易避免，我在评论区发了图片，你可以参考一下。

@bilibilissr 😄

There are several models aiming on this purpose, I have no such plan at least for now.😊

Sorry for late. This is an embedding, a negative embedding. You should type "embedding:NegfeetV2.pt" or "NegfeetV2" inside negative prompt bar depends on what UI you're using.

Yes. But now the tools have not developed and trainning method is unclear for me. I will working on this at right time.

No, this embedding is not for this purpose. Inpainting can do this well.

Of course, but the quality may differ between checkpoints.

你好。这取决于你用的是什么启动器，一般来讲是Comfy或者A1111，对于这两者而言都是放在其中的“embeddings”文件夹中，如果是用的其他启动器，大概其结构也是如此。如果你要使用其它类型的模型，放在特定的文件夹中就行了，例如lora模型放在"lora"文件夹中。

Hi. Which UI you're using? A1111 or comfy or something else? The right way for you is to check the website or readme documentation of the software you are using. I think most people use A1111, if so as you, you need to check the generation data below the preview, if you see the "embedding hash", so it works.

No, this is not for pony or sdxl models.

大概率是不会的，如果加在正向里面好的话有两个可能：第一是用了该嵌套本来就不支持的checkpoint比如sdxl，导致嵌套没有正常加载，而只是作为文本被text encoder编码了；第二是用的checkpoint的text encoder相当程度上被魔改了，导致嵌套的功能发生了很大变化。

You mean to update to sdxl or to update for newer sd1.5 model? Maybe but right now I'm on training my Barphot model, a hard task but a big deal.

@eoa42 yes after Barphot is done, I'll try train an NegfeetXL

@eoa42 Sorry for late. I had tested negtive embeddings on SDXL last year, compared to SD1.5 it didn't show as much help for feet, so I dropped this plan. Special text encoder structure might be the reason. For now I'm on Zimage training.