Using the SDXL controlnet framework to simulate the "Diffusers Image Fill" guide, this workflow provides a solution for object removal.

In the past, when attempting to remove objects, especially people, the result often ended up being replaced by different individuals, making removal impossible. Additionally, the removed areas would sometimes be filled with blurry artifacts or strange shapes. These issues are usually caused by problems with the mask.  

In this workflow, the SAM model is used to identify and segment the object to be removed, and the mask is expanded slightly. By referring to the "Diffusers Image Fill" guide, the mask is processed in a black-and-white format. Expanding the mask is a critical step. If the subject you want to remove keeps being replaced with other objects, expanding the mask is necessary to achieve successful removal.  

The Realvis model is used in this workflow, and the required URL is provided in the notes. The SAM model can be downloaded automatically. Additionally, the workflow allows for manual mask drawing—simply enable the purple group's Boolean setting by setting it to "True."  

==

使用SDXL模擬"Diffusers Image Fill" guide以建立的工作流。

過去在移除物體時，往往要移除人物，卻會變成別的人物，怎樣也無法移除，此外，移除區域也常會被填上模糊的怪物，實際上這都是遮罩的問題。

在工作流中，使用SAM模型指定分割的物體，並簡單的擴張遮罩，參考"Diffusers Image Fill" guide的指南內容，將遮罩採黑白分色的方式處理，其中，擴張遮罩是重要的關鍵，若你要消除的主體不斷的被變為別的物體，則必須擴張遮罩以完成移除。

模型的部分使用Realvis，在工作流的Note中有提示所需的網址，SAM的部分可自動下載，此外，工作流也可以使用手繪遮罩，只要打開紫色群組的Boolean將它設置為True即可。