A specialized masking tool designed for ComfyUI, this utility enables users to individually segment and extract detailed contours of multiple characters within a single image. It addresses the limitations of traditional bounding box methods, offering refined control for multi-person scenarios such as inpainting and face swapping.
- Provides two nodes, EveryPersonSegSimple and EveryPersonSegDetail, for various levels of segmentation detail.
- Utilizes advanced YOLOv8 and MediaPipe models for accurate and fine-grained portrait masking.
- Allows customization of output masks through parameters like confidence thresholds and sorting methods.
Context
This tool, known as ComfyUI-Every-Person-Seg-CoiiNode, enhances the ComfyUI framework by offering capabilities to break down the intricate contours of characters in images. Its primary function is to facilitate multi-person generation workflows, which are increasingly demanded in creative applications.
Key Features & Benefits
The tool features two main nodes: EveryPersonSegSimple for basic segmentation and EveryPersonSegDetail for advanced, detailed masking. This flexibility allows users to choose the level of detail they need, making it suitable for various use cases, from simple edits to complex multi-person compositions.
Advanced Functionalities
EveryPersonSegDetail leverages both YOLOv8 and MediaPipe models to produce highly detailed masks that include different parts of a person, such as the face, hair, body, and clothing. This node also incorporates an option for mask contour optimization, enhancing the quality of the segmentation and ensuring that each mask accurately reflects the contours of the subjects.
Practical Benefits
By providing precise segmentation capabilities, this tool significantly improves the workflow in ComfyUI for tasks like inpainting and face swapping. Users gain greater control over individual character elements, leading to higher quality outcomes and more efficient editing processes.
Credits/Acknowledgments
The project acknowledges the contributions of the original authors, particularly for the contour optimization code derived from the djbielejeski/a-person-mask-generator. The tool is developed under an open-source license, promoting community collaboration and enhancement.