search

TencentARC / BrushNet

[ECCV 2024] The official implementation of paper "BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion"
https://tencentarc.github.io/BrushNet/
Other
1.15k stars 97 forks source link

SDXL checkpoint bad result? #43

Open lingcong-k opened 3 weeks ago

lingcong-k commented 3 weeks ago

Describe the bug

I use test_brushnet_sdxl.py and use the segmentation sdxl checkpoint on the cake images in the example folder. but i got very bad results. am I doing sth wrong? "A cake on the table." image image

result image

Reproduction

run test_brushnet_sdxl.py. but with stable-diffusion-xl-base-1.0 checkpoint and segmentation_mask_brushnet_ckpt_sdxl_v0

Logs

No response

System Info

A100

Who can help?

No response

Shuvo001 commented 2 weeks ago

@lingcong-k you can try to generate the outputs with ComfyUI. SDXL version outputs are not as bad as yours

lawsonxwl commented 1 week ago

为什么用 test_brushnet_sdxl.py 输出的是inpaint背景的结果？ 我这边用这个代码和图片输出的是inpaint前景的结果。。。