In your paper, you mention that the reference image is inpainted using SD-inpainting with the prompt “background.” However, when I use "background" as the inpainting prompt, the fooocus model often generates unwanted objects. Could you provide more details on the reference inpainting process?

