This repository provides advanced nodes for ComfyUI, extending the functionality of the original ComfyUI Layer Style. It targets users who need more complex image manipulation and AI-driven processing capabilities within ComfyUI, offering a wide array of specialized nodes for tasks like advanced image editing, object detection, and detailed segmentation.

How It Works

The project offers a collection of custom nodes that integrate various AI models and image processing techniques directly into the ComfyUI workflow. It leverages pre-trained models for tasks such as object detection (YOLO, Florence2, Gemini), segmentation (SAM, BiRefNet, Mediapipe), and text/image generation (Gemini, DeepSeek, ZhipuGLM, Phi-3.5, Llama). The nodes are designed to be modular, allowing users to chain them together for complex pipelines.

Quick Start & Requirements

• Installation: Recommended via ComfyUI Manager. Alternatively, clone the repository into ComfyUI/custom_nodes and run install_requirements.bat (or install_requirements_aki.bat for Aki ComfyUI).
• Dependencies: Python, PyTorch, CUDA (for GPU acceleration), and various AI model libraries. Specific model files need to be downloaded and placed in the ComfyUI/models directory.
• Resources: Requires significant VRAM for advanced models (e.g., 16GB for Phi-3.5 vision models). Model downloads can be substantial.

Highlighted Details

• Extensive support for multiple LLM and Vision-Language Models (Gemini, DeepSeek, ZhipuGLM, Phi-3.5, Llama, SmolLM, etc.).
• Advanced image segmentation and background removal with various models (SAM, BiRefNet, TransparentBackground, PersonMaskUltra).
• Sophisticated object detection and bounding box manipulation tools.
• Includes nodes for image collage, prompt generation, and image editing.

Maintenance & Community

The project is actively maintained, with recent commits adding support for new models and features like Gemini 2.0, DeepSeek API V2, and improved SAM implementations. Links to community resources are not explicitly provided in the README.

Licensing & Compatibility

The nodes follow the MIT license. However, the README states that if used for commercial purposes, users must refer to the original project licenses for any functional code derived from other open-source projects.

Limitations & Caveats

• Some nodes may have specific dependency requirements that need manual installation (e.g., psd_tools, yolo-world).
• The README notes potential issues with specific package versions (e.g., opencv-contrib-python, transformers, protobuf) and provides troubleshooting steps.
• Images larger than 2K using VITMatte edge processing may consume significant memory.