This repository provides an unofficial implementation of YOLO-World and EfficientSAM for ComfyUI, enabling efficient object detection and instance segmentation. It targets users of ComfyUI, particularly those involved in image and video processing, offering advanced mask manipulation capabilities.

How It Works

The integration leverages YOLO-World for object detection and EfficientSAM for precise instance segmentation. Users can select specific YOLO-World models (l, m, s) and EfficientSAM models, configuring parameters like confidence and IoU thresholds. A key feature is the ability to either merge all segmentation masks into a single output or extract specific masks based on their index, supporting both image and video workflows.

Quick Start & Requirements

• Install: cd custom_nodes && git clone https://github.com/ZHO-ZHO-ZHO/ComfyUI-YoloWorld-EfficientSAM && cd ComfyUI-YoloWorld-EfficientSAM && pip install -r requirements.txt
• Prerequisites: Python, ComfyUI. EfficientSAM models (efficient_sam_s_cpu.jit, efficient_sam_s_gpu.jit) must be manually downloaded and placed in the custom node directory. GPU acceleration is supported.
• Resources: Requires downloading model weights. Setup time is minimal after cloning and installing requirements.
• Links: V2.0 Video Demo

Highlighted Details

• Supports YOLO-World models (l, m, s) and EfficientSAM.
• Advanced mask handling: merge all masks or extract specific ones.
• V2.0 adds mask separation and extraction for images and videos.
• Includes a Yoloworld ESAM Detector Provider node, compatible with Impact-Pack.

Maintenance & Community

• Active development with V2.0 released recently (Feb 2024).
• Contact via email (zhozho3965@gmail.com) or QQ group (839821928).
• Links to Bilibili, X (Twitter), and Xiaohongshu provided.

Licensing & Compatibility

• License not explicitly stated in the README.
• Compatibility with commercial or closed-source projects is undetermined.

Limitations & Caveats

The project is an unofficial implementation. The V1.0 workflow is deprecated and incompatible with V2.0. Manual model downloading is required, and licensing for commercial use is not specified.