BrushNet

BrushNet is a novel image inpainting model hat focuses on preserving image details and structures while filling in missing or masked regions. Unlike traditional inpainting methods that may distort original content, BrushNet ensures high-fidelity reconstruction by incorporating multi-stage fusion and reference feature injection.

Version 1.34.1 introduces the BrushNet option for inpainting, currently available exclusively for SDXL models.

For subtle image edits, it's better to use low-strength inpainting (around 0.4-0.6) without BrushNet, as BrushNet tends to make more noticeable changes to the inpainted area.

Check the box next to BrushNet to enable this BrushNet feature.

Let’s take a look at the examples below to see the difference between inpainting with and without BrushNet.

In this example, we’ll add a beer to the table in the photo by masking the region for inpainting.

Prompt: A beer on the table

Let's see the result with and without BrushNet:

With BrushNet

Wihtout BrushNet, 0.8 strength

The results clearly show that BrushNet preserves the original image structure while seamlessly adding the beer to the table.