ComfyOnline
Easy automatic (square) image cropper using Yolo

ComfyUI-Yolo-Cropper

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 1024x1024 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

image

Install

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

1. goto ComfyUI/custom_nodes dir in terminal(cmd)
2. git clone https://github.com/tooldigital/ComfyUI-Yolo-Cropper.git
3. Restart ComfyUI

Input

  • 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)

Output

  • 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

Basic flow with face and person

image

Basic flow with object - parking meter

image

Flow using IP-Adapter FaceID

image