ComfyUI-Yolo-Cropper

★ 15

目标检测方形裁剪自动下载模型ControlNet/IP-adapter兼容

基于YoloV8自动检测并生成方形图像裁剪与掩码，支持人脸检测并自动下载模型，便于ControlNet/IP-adapter等1:1输入场景。

💡 为ControlNet或IP-adapter快速生成1:1方形裁剪与掩码。

🍴 6 Forks💻 Python🔄 2024-06-14

📦

网盘下载

复制链接后前往夸克网盘下载

https://pan.quark.cn/s/79aaff81621b

📄 README

A very simple ComfyUI node to help you create image crops and masks using YoloV8.

It is usefull for creating square crops when working with controlnets, ip-adapters etc that need 1:1 ratio input like 1024×1024 for example.

It uses YoloV8 – Ultralytics – with COCO 80 objects for detection added with a specific feature for face detection.

When multiple objects are detected, it will return the biggest bounding box.

All necessary models will be downloaded automatically at first use.

https://github.com/ultralytics/assets/releases/download/v8.2.0/yolov8l.pt

https://huggingface.co/spaces/cc1234/stashface/resolve/main/.deepface/weights/yolov8n-face.pt

To install this node in addition to an existing installation of ComfyUI, you can follow the following steps:

goto ComfyUI/custom_nodes dir in terminal(cmd)

git clone https://github.com/tooldigital/ComfyUI-Yolo-Cropper.git

Restart ComfyUI

image = Any input image will do

object = dropdown list with COCO 80 objects, it has face as an extra.

padding = padding that is added to the square crop (not the orginal crop)

bounding box = array of values x,y,width,height

image yolo detections = an image showing all detected assets and bounding boxes

image original crop = the image cropped according the actual bounding box

image square crop = the image cropped according a square bounding box starting from the center of the original bounding box

mask orginal crop = a black and white mask cropped according the actual bounding box

mask square crop = a black and white mask cropped according a square bounding box starting from the center of the original bounding box