This repository offers a set of custom nodes for ComfyUI that enable anime-style image segmentation using optimized pre-trained models. It features two primary nodes: SimpleAnimeSeg for quick segmentation and AdvancedAnimeSeg for more detailed, instance-based character segmentation.
- Efficient CPU-based segmentation, with potential future GPU support.
- SimpleAnimeSeg provides rapid results, while AdvancedAnimeSeg allows for customizable threshold adjustments.
- Both nodes utilize ONNX models for enhanced performance and integration.
Context
The ComfyUI Anime Segmentation Nodes are designed to enhance the capabilities of ComfyUI by providing specialized tools for segmenting anime-style images. Their primary purpose is to facilitate the extraction of characters from backgrounds, making them valuable for artists and developers working with anime-related content.
Key Features & Benefits
The tool includes two distinct nodes: SimpleAnimeSeg and AdvancedAnimeSeg. SimpleAnimeSeg is optimized for speed, making it suitable for scenarios where characters are easily distinguishable from their backgrounds. In contrast, AdvancedAnimeSeg offers a more granular approach, allowing users to adjust thresholds for better accuracy in complex scenes.
Advanced Functionalities
AdvancedAnimeSeg stands out with its multi-stage processing approach, combining detection and mask refinement to produce cleaner outputs. This node utilizes advanced techniques from the CartoonSegmentation project, enhancing its ability to manage intricate character details and backgrounds. Users can leverage its flexibility to achieve more precise segmentation results.
Practical Benefits
By integrating these nodes into their workflow, users can significantly improve their efficiency and control over the segmentation process in ComfyUI. The ability to quickly and accurately isolate characters not only saves time but also enhances the quality of final outputs, allowing for better downstream editing and creative possibilities.
Credits/Acknowledgments
This project acknowledges the contributions of several key figures, including SkyTNT for the ONNX segmentation model, city96 for the original segmentation node, and dreMaz for providing advanced model weights. The framework is built upon the ComfyUI platform, which serves as the foundation for these custom nodes.