SUPIR Model Loader (v2) Node Documentation

Overview

The SUPIR Model Loader (v2) node is a component of the ComfyUI SUPIR upscaler wrapper, designed to efficiently manage the loading of SUPIR models used for image upscaling and restoration tasks. This node forms part of a collection of nodes that decompose the SUPIR process into manageable sections, allowing for greater flexibility and understanding of the underlying mechanics.

Node Functionality

What This Node Does

The SUPIR Model Loader (v2) node is responsible for loading the required SUPIR model checkpoints, which are used in conjunction with other nodes to perform advanced image restoration and upscaling. It utilizes pruned models in safetensors format, enabling efficient memory usage and faster model loading, making it a pivotal component in setting up workflows within ComfyUI that involve SUPIR processing.

Inputs

The SUPIR Model Loader (v2) node requires specific inputs to function correctly:

1. Model Checkpoint Path: The path to the SUPIR model(s), which should be stored in the ComfyUI/models/checkpoints directory. Users need to ensure that they have the appropriate model versions available for loading.

2. SDXL Model Availability: This node expects an SDXL model to be available in the same directory as the SUPIR models.

3. Additional Inputs: If applicable, inputs such as prompts or specific configurations may be provided, depending on the workflow requirements involving further nodes.

Outputs

This node does not produce immediate outputs that can be visualized, as its primary function is to load models that will be utilized by other downstream nodes in the workflow. However, the successful loading of models is crucial for the smooth execution of image processing operations in subsequent nodes.

Usage in ComfyUI Workflows

The SUPIR Model Loader (v2) node is an integral part of workflows aiming to leverage the SUPIR upscaling capabilities. It is typically used at the beginning of a process pipeline, ensuring that all necessary model files are present and loaded before any image processing begins. By using this node, users can:

• Ensure efficient and correct loading of SUPIR models in safetensors format.
• Seamlessly integrate with other ComfyUI nodes designed for image denoising, encoding, decoding, and upscaling.
• Facilitate improved hardware support and model compatibility, crucial for memory-intensive tasks like high-resolution upscaling.

Special Features and Considerations

• Memory Efficiency: The node’s use of pruned safetensors format ensures reduced memory footprint, allowing for large-scale image processing on GPU hardware with limited VRAM.

• Automatic Detection: The node automatically detects the presence of xformers, which can offer performance improvements in specific cases, optimizing model loading times.

• Scalability: The node supports a variety of input resolutions and scales efficiently according to hardware capabilities, making it versatile for different use cases and GPU configurations.

• Non-Commercial Usage: Users should be aware that the SUPIR framework is intended for non-commercial purposes only, ensuring compliance with usage restrictions outlined in the licensing agreement.

• Integration with LoRAs: The improved model loading strategy allows for compatibility with LoRAs, enriching the flexibility and potential utility in complex image processing scenarios.

In summary, the SUPIR Model Loader (v2) node serves as a foundational component in ComfyUI SUPIR workflows, managing the crucial task of model loading with an emphasis on efficiency, flexibility, and compliance with usage terms.