ComfyUI-NuA-BIRD

ComfyUI-NuA-BIRD
★ 8

盲图像修复非均匀去模糊inpainting 与修补超分辨率
在ComfyUI中实现BIRD盲图像修复，利用快速扩散反演完成去模糊、去噪、inpainting与超分，支持非均匀模糊，能在无精确退化模型下恢复图像细节。
💡 对模糊、噪声或缺损图像进行一键恢复与超分处理。
🍴 3 Forks💻 Python🔄 2024-06-18
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/79aaff81621b
📦 requirements.txt
numpy
torch
torchvision
blobfile
tqdm
pyYaml
pillow
diffusers
📄 README

ComfyUI-NuA-BIRD

ComfyUI implementation of “Blind Image Restoration via Fast Diffusion Inversion”

Original article

Features

  • Blind Deblurring
  • Non-uniform Deblurring
  • Inpainting
  • Denoising
  • Superresolution

    • Installation

  • Clone the repository into the ComfyUI/custom_nodes directory

    • “`sh

    cd ComfyUI/custom_nodes

    git clone https://github.com/nuanarchy/ComfyUI-NuA-BIRD.git

    “`

  • Install the required modules

    • “`sh

    pip install -r ComfyUI-NuA-BIRD/requirements.txt

    “`

  • Copy the model weights into the appropriate folder

    • ComfyUI/models/checkpoints

    Examples

    In the examples folder, you will find the workflow diagrams, the JSON file with the configurations, and resulting images.

    Workflow Diagrams

    Blind Deblurring

    Non-uniform Deblurring

    Inpainting

    Denoising

    Super Resolution

    Important

    The results primarily depend on the pretrained model and the dataset

    Limitations:

  • The model only works with square images at a resolution of 256×256 pixels
  • Faces must be cropped and centered in the images
  • For Super Resolution tasks, the input image resolution can be any size smaller than 256×256 pixels

    • If you want to overcome these limitations, you can train your own diffusion model using custom datasets.

    You can use the OpenAI repository: improved-diffusion