Issues with the image relighting model

[jayLeo456]

Hello, I have a question regarding the implementation in the CLA module, the current implementation takes the consistent target as input, whereas the official IC-Light uses the image foreground as input. Would it be feasible to first apply foreground segmentation to the consistent target and use the segmented result as input instead?

[YujieOuO]

https://github.com/bcmi/Light-A-Video/blob/main/lav_paint.py#L211
Are you referring to the background inpainting task of Light-A-Video? In this task, we first segment the foreground and then use the background generation capability of Animatediff to repaint and relight simultaneously.