Compositing Settings

The Compositing Settings are used to join an inpainted or outpainted section of image with the original image. This ensures that newly generated parts of the image integrate seamlessly with previously existing parts of the image.

TABLE OF CONTENTS

• Coherence Pass
• Infill

Coherence Pass

The Coherence Pass settings determine how the denoising process will be performed on the edges of the seams between the new generation and the old generation.

Method

What part of the image will have the the Compositing applied to it.

• Gaussian Blur will perform a Gaussian blur to determine the edged area to perform gradient denoising over.
• Box Blur will perform a Box blur to determine the edged area to perform gradient denoising over. 
• Staged will progressively increase the denoising area in a systematic way -- It is recommended to ensure that a minimum denoising value is set for this method.

Edge Size

• The size of the Edge to perform Gradient Denoising over

Min Denoise

• The minimum amount of Denoising to perform at any point on the masked area - This can avoid areas where insufficient denoising is performed - Increase this if you see any artifacts on the edges of your generation. When in doubt, set this to .1-.2 if you're finding artifacts in your inpainting.

Mask Blur

• The blur to apply on the mask before performing denoising and in compositing the original/new image.

Infill

Infill Method

Method to infill the selected area. Different methods function differently and will produce different result.

• tile uses tiling to infill
• patchmatch uses the PatchMatch algorithm
• lama uses the LaMa method of infilling
• cv2 uses OpenCV to infill

Below, you can see how the infill method creates new content in the transparent area (top row), and how that impacts the resulting generation. We typically recommend Patchmatch as a solid, multi-purpose infill method.