This repository offers a custom node for ComfyUI that leverages TensorRT to significantly accelerate depth map generation, achieving speeds up to 14 times faster than existing solutions. It supports multiple model versions, including Depth Anything v1, v2, and Distill Any Depth, enhancing the capabilities of depth estimation in AI art workflows.
- Utilizes TensorRT for optimized performance, enabling real-time depth map generation.
- Supports various model configurations, allowing users to choose based on their specific needs and hardware capabilities.
- Offers a user-friendly node integration within ComfyUI, simplifying the workflow for artists and developers.
Context
The ComfyUI Depth Anything TensorRT extension is designed to enhance the depth mapping capabilities within the ComfyUI framework. By implementing TensorRT, it allows users to generate depth maps more efficiently, making it a valuable tool for artists working with AI-generated imagery.
Key Features & Benefits
This tool provides ultra-fast depth map generation, which is crucial for creating high-quality AI art. The ability to select from different model versions means users can optimize their performance based on their hardware, ensuring that they can work effectively without compromising quality.
Advanced Functionalities
The extension includes advanced options for building TensorRT engines, either automatically through a dedicated node or manually by downloading ONNX models. This flexibility allows users to tailor their setup according to their specific requirements, facilitating a more customized workflow.
Practical Benefits
The integration of this tool into ComfyUI streamlines the workflow for generating depth maps, significantly improving efficiency and control over the output quality. Users can expect faster processing times, enabling more iterations and experimentation in their creative processes.
Credits/Acknowledgments
The development of this tool acknowledges contributions from various authors and repositories, including NVIDIA's Stable Diffusion TensorRT implementation and other related projects. The license for this repository is MIT, allowing for wide usage and modification within the open-source community.