这个工作流主要功能就是把源图像的脸放到目标图像上并让它们融合，这是一种覆盖而不是重绘。它可能融合效果不好，但是一定像，因为这本就是原来的脸。

图像的选择：

1.目标图像需要尽量与源图像保持相同的角度，这很重要，我无法通过任何方法解决角度不同的问题。

2.图像尽量不要选择多人物的，这样会让检测节点失效，尽管我已经尽量避免这方面的错误。

PS:这其中我想还有许多的改进空间，比如边缘重绘部分的遮罩宽度，我在过去经常使用的方法是利用yolo裁剪的面部box的边长做线性的公式计算来稳定这个宽度，但是也常常有错漏，所以这次我没那么做。这会导致一些小问题，但也许能让下载并使用我的工作流的人能够好好研究下它而不是依赖全自动而完全不去思考，没有任何贡献，我可不是什么赛博菩萨。

还有很多很多的问题，我就不一一展开了，欢迎大家讨论，我需要大家一同把这个工作流做得更完善。

Sorry, my English is not good, this comment is from web translation.

The main function of this workflow is to place the face of the source image into the target image (this is an overlay, not a redraw). After this, the edges of the stitching are redrawn, so that the target image has a face of the source image.It may not blend well, but it must be like, because this is the original face.

Image Selection:

1. The target image needs to be at the same angle as the source image as much as possible, which is important, and I can't solve the problem of different angles by any method.

2. Try not to select multiple characters in the image, as this will invalidate the detection node, although I have tried to avoid errors in this regard.

Among them, I think there are many improvements, such as the mask width of the edge redrawing part. The method I used to use in the past was to use the edge length of the face box cropped by yolo to do a linear formula calculation to stabilize this width, but there are often mistakes, so I did not do it this time.This can lead to some minor issues, but it may allow people who download and use my workflow to study it well instead of relying on full automation and not thinking at all, without any contribution, I am not a bodhisattva.

There are still a lot of questions, I will not open them together, welcome everyone to discuss, I need you to work together to improve this workflow.