Workflow: DiffBIR High-definition Facial Enhancement with 8GB of Operational Capacity

DiffBIR high-definition enlargement, facial repair, image denoising

Project address:

https://github.com/ComfyUI/ComfyUI-DiffBIR

Installation:

1. cd custom_nodes
2. git clone https://github.com/ComfyUI/ComfyUI-DiffBIR
3. pip install -r requirements.txt
4. Restart ComfyUI

Download the model to: ComfyUI/models/diffbir/

All image download links are below:

https://kkgithub.com/cszn/KAIR/releases/download/v1.0/BSRNet.pth

https://hf-mirror.com/lxq007/DiffBIR/resolve/main/face_swinir_v1.ckpt

https://kkgithub.com/cszn/KAIR/releases/download/v1.0/scunet_color_real_psnr.pth

https://hf-mirror.com/lxq007/DiffBIR/resolve/main/general_swinir_v1.ckpt

https://hf-mirror.com/lxq007/DiffBIR-v2/resolve/main/v2.pth

https://hf-mirror.com/stabilityai/stable-diffusion-2-1-base/resolve/main/v2-1_512-ema-pruned.ckpt

Tip: When Tiled is enabled, it will take a little longer, and without it the speed is much faster

     4090 24g card, float32 4x enlargement, when using, pay attention to calling clearGPU

     If the computing power is not very good, it is not recommended to try float32, the effect is not much better than float16