comfyui-every-person-seg-coii

ComfyUI-Every-Person-Seg-CoiiNode

English|中文

Update

• 2026-3-6: Added gender-based sorting support (female-male, male-female) using OpenCV DNN gender detection. Added combine option to merge all detected person masks into one.
• 2025-6-20: Initial release

Project Introduction

A tool that provides the ability to split fine-grained contours of people one by one for multi-person usage scenarios.

Development Reason

There is a demand for multi-person generation in ComfyUI applications, but many tools for segmenting person IDs use bbox, which has the problem of not being able to separate multiple people cleanly. Therefore, this custom node plugin was created. It provides necessary implementation capabilities for scenarios such as multi-person loop inpaint and face swapping.

ComfyUI Node Description

EveryPersonSegSimple Node

EveryPersonSegSimple Node with person_fullfil:True

EveryPersonSegDetail Node

EveryPersonSegSimple

Input Parameters

• images: Input image
• yolov_path: YOLOv8 segmentation model path (default uses person_yolov8m-seg.pt)
• confidence: Detection confidence threshold (0.0 ~ 1.0)
• drop_area: Minimum reserved area percentage (0 ~ 99)
• person_fullfil: Whether to generate complete portrait neighboring fill area (person_area) or original portrait individual mask (person_masks)
• order: Mask sorting method (options: confidence, large-small, small-large, left-right, right-left, top-bottom, bottom-top, female-male, male-female)
• combine: Whether to merge all detected person masks into one mask (enabled/disabled)

Output

• mask: Portrait mask collection

Usage Example

Drag the workflow example/workflowexample_everypersonseg.json into ComfyUI to use:

EveryPersonSegSimple And EveryPersonSegDetail

Dependencies

• mediapipe>=0.10.13
• ultralytics>=8.0.0
• opencv-python>=4.5.0
• Pillow>=9.0.0
• scipy>=1.7.0

Manual Model Download

If automatic download fails, please manually download the following models and place them in the specified paths:

YOLOv8 Segmentation Model

• Download URL: https://huggingface.co/Bingsu/adetailer/resolve/main/person_yolov8m-seg.pt
• Save Path: ComfyUI/models/ultralytics/segm/person_yolov8m-seg.pt

MediaPipe Portrait Segmentation Model

• Download URL: https://huggingface.co/yolain/selfie_multiclass_256x256/resolve/main/selfie_multiclass_256x256.tflite?download=true
• Save Path: ComfyUI/models/mediapipe/selfie_multiclass_256x256.tflite

OpenCV Gender Detection Models

• Gender Deploy Prototxt: https://raw.githubusercontent.com/smahesh29/Gender-and-Age-Detection/master/gender_deploy.prototxt
• Gender Net Caffemodel: https://raw.githubusercontent.com/smahesh29/Gender-and-Age-Detection/master/gender_net.caffemodel
• Haar Cascade: https://raw.githubusercontent.com/opencv/opencv/master/data/haarcascades/haarcascade_frontalface_default.xml
• Save Path: ComfyUI/models/opencv_dnn/

Acknowledgments

EveryPersonSegDetail

This node combines YOLOv8 segmentation model and MediaPipe portrait segmentation model to provide more detailed portrait masks (supporting fine segmentation of face, hair, body, clothes and other parts). The mask contour optimization part of the code refers to the djbielejeski/a-person-mask-generator project.

Input Parameters

• images: Input image
• face_mask/hair_mask/body_mask/clothes_mask/background_mask: Select mask types to generate
• refine_mask: Whether to enable mask contour optimization
• confidence: Detection confidence threshold
• yolov_path: YOLOv8 segmentation model path
• drop_area: Minimum reserved area percentage

Output

• person_masks: Refined portrait part mask collection

---

Acknowledgments

The mask contour optimization part of the EveryPersonSegDetail node in this project is modified and implemented based on the djbielejeski/a-person-mask-generator project.