ComfyUI Node: FreeU

FreeU:

FreeU is a node designed to enhance the performance and quality of AI-generated images by applying advanced filtering and scaling techniques. It leverages the power of Fourier filtering and dynamic scaling based on the model's channel configuration to refine the output. The primary goal of FreeU is to improve the visual fidelity and detail of images produced by AI models, making it a valuable tool for AI artists looking to achieve higher quality results. By dynamically adjusting the hidden states and applying Fourier filters, FreeU ensures that the generated images are not only visually appealing but also maintain a high level of detail and clarity.

FreeU Input Parameters:

model_channels

This parameter represents the number of channels in the model's configuration. It is used to determine the scaling factors for the hidden states and the Fourier filter. The value of model_channels directly impacts the scaling dictionary, which in turn affects the intensity and application of the filters. The exact value is derived from the model's UNet configuration and is crucial for the proper functioning of the node.

scale_dict

The scale_dict parameter is a dictionary that maps the model channels to their respective scaling factors. It contains pairs of values that determine how the hidden states and Fourier filters are scaled. The dictionary is constructed using the model channels and predefined scaling factors, ensuring that the filters are applied correctly based on the model's configuration. This parameter is essential for dynamically adjusting the filters to match the model's architecture.

on_cpu_devices

This parameter is a dictionary that keeps track of devices that do not support certain torch.fft functions. If a device is found to be incompatible, the Fourier filter operations are switched to the CPU. This ensures that the node can function correctly even on devices with limited support for specific operations. The on_cpu_devices dictionary helps maintain compatibility and stability across different hardware configurations.

FreeU Output Parameters:

h

The h parameter represents the modified hidden states after applying the scaling factors. It is an intermediate output that reflects the adjustments made to the hidden states based on the scaling dictionary. The h parameter is crucial for ensuring that the hidden states are appropriately scaled, leading to improved image quality.

hsp

The hsp parameter represents the modified hidden states after applying the Fourier filter. It is an intermediate output that reflects the adjustments made to the hidden states based on the Fourier filter and the scaling factors. The hsp parameter is essential for ensuring that the hidden states are appropriately filtered, leading to enhanced image detail and clarity.

FreeU Usage Tips:

• Ensure that your model's configuration is correctly set up, as the model_channels parameter is derived from the model's UNet configuration.
• Use appropriate scaling factors in the scale_dict to match the model's architecture and achieve the desired filtering effects.
• If you encounter issues with certain devices not supporting torch.fft functions, consider switching to CPU for those operations to maintain compatibility.

FreeU Common Errors and Solutions:

Device does not support the torch.fft functions used in the FreeU node, switching to CPU.

• Explanation: This error occurs when the device being used does not support the required torch.fft functions for Fourier filtering.
• Solution: The node automatically switches to CPU for the Fourier filter operations. Ensure that your system has sufficient CPU resources to handle the operations, or consider using a different device that supports the required functions.