I have converted the sdxl inpainting checkpoint to the original sdxl implementation using the script convert_diffusers_to_original_sdxl.py. When I try to run inference with it by using GitHub - Stability-AI/generative-models: Generative Models by Stability AI script ( after changing the implementation to support inpainting), it gives me samples with a dark lighting. Does anyone have an idea why this is happening? The uploaded pictures are showing the original picture and the picture after inpainting
