When I combined Stand-In with Vace, use Vace's inpainting ability to replace the person in the video, I met some problems.
first, I use flux kontext to replace the person in the first frame of video with ip_image to create reference_image, then I create a mask video in which white region indicts the region of person and black region indicts the region of background, next I set person's region of video to gray like Vace's preprocess do, and finally I pass ip_image、reference_video、mask_video、reference_image to infer_with_vace (I simply modified it to add mask_video input).
But I found the first frames of the video is weired, it feels like some noise has been superimposed in person's region and this problem exists in multiple test cases.
Do you know why and how to fix this problem?
When I combined Stand-In with Vace, use Vace's inpainting ability to replace the person in the video, I met some problems.
first, I use flux kontext to replace the person in the first frame of video with ip_image to create reference_image, then I create a mask video in which white region indicts the region of person and black region indicts the region of background, next I set person's region of video to gray like Vace's preprocess do, and finally I pass ip_image、reference_video、mask_video、reference_image to infer_with_vace (I simply modified it to add mask_video input).
But I found the first frames of the video is weired, it feels like some noise has been superimposed in person's region and this problem exists in multiple test cases.
Do you know why and how to fix this problem?