这是一个通过选择颜色来对图像进行抠图的案例，原理是先用segment将图片中的元素用不同颜色进行分割，接下来使用吸管工具来获取颜色，最后通过蒙版来完成最终的抠图，最终得到透明背景的图片。在抠图相率上，比segment anything v2差一点点，操控性上个人认为要高一些，毕竟有些人对“抱枕”和“靠枕”的概念有些模糊，当然这都是语义识别模型的问题了。

用法：

1、上传图片

2、执行队列

3、利用取色器中的吸管获取seg中的颜色

4、再次执行队列即可

This is a case of cutting out an image by selecting a color. The principle is to first use segments to segment the elements in the image with different colors, then use the eyedropper tool to get the color, and finally use a mask to complete the final cutout. , and finally get a picture with a transparent background. In terms of cutout ratio, it is a little worse than segment anything v2. In terms of controllability, I personally think it is higher. After all, some people are a little vague about the concepts of "cushion" and "back pillow". Of course, this is all a problem of the semantic recognition model. .

usage:

1. Upload pictures

2. Execution queue

3. Use the straw in the color picker to get the color in the seg.

4. Execute the queue again

plug-in： https://github.com/vxinhao/comfyui-minitools

video: https://www.bilibili.com/video/BV1y742ewEog/?share_source=copy_web&vd_source=510b17607accdb1e808d71173429a19d