Dose it support ControlNet inpaint in SDXL？

[ShenZhang-Shin]

Yes, I'm working on it. It works fine. I will integrate it into the code soon. Here is an example. I directly edit 2K image using controlnet inpainting + hidiffusion input:

output:

[Whoo-jl]

Yes, I'm working on it. It works fine. I will integrate it into the code soon.是的，我正在努力。它工作正常。我将很快将其集成到代码中。 Here is an example. I directly edit 2K image using controlnet inpainting + hidiffusion下面是一个示例。我使用 controlnet inpainting + hidiffusion 直接编辑 2K 图像 input: 输入：

output: 输出：

That's great! I can't wait to try it!

[Yzhbuaa]

Hello, which inpainting controlnet for SDXL did you use in the example above?

[ShenZhang-Shin]

Hello, which inpainting controlnet for SDXL did you use in the example above?

diffusers/controlnet-canny-sdxl-1.0

[Yzhbuaa]

Thank you for your quick reply! Did you mean that you use stabilityai/stable-diffusion-xl-base-1.0 conbined with diffusers/controlnet-canny-sdxl-1.0 to perform the image inpainting task? What is the benefit of using a canny controlnet rather than only sdxl?

[ShenZhang-Shin]

Thank you for your quick reply! Did you mean that you use stabilityai/stable-diffusion-xl-base-1.0 conbined with diffusers/controlnet-canny-sdxl-1.0 to perform the image inpainting task? What is the benefit of using a canny controlnet rather than only sdxl?

Yes.
Only use stabilityai/stable-diffusion-xl-base-1.0 can also do inpainting task, but the pose consistency is not as good as sdxl+controlnet. For example, here is the inpainting output only using sdxl+hidiffusion. You can see the difference. The inpainting code in readme is sdxl+hidiffusion. I will integrate sdxl+controlnet+hidiffusion into our codebase.