ComfyUI Node: Flip Sigmas Adjusted

Flip Sigmas Adjusted:

The FlipSigmasAdjusted node is designed to manipulate a tensor of sigma values, which are often used in various AI and machine learning processes, particularly in noise scheduling for diffusion models. This node flips the order of the sigma values and then adjusts them based on specified parameters. The adjustments include offsetting the sigma values by a given index, ensuring no zero values are present, and optionally normalizing the values by the last sigma or a specified divisor. This node is particularly useful for AI artists who need to fine-tune the noise schedules in their models, providing more control over the diffusion process and potentially improving the quality of generated images.

Flip Sigmas Adjusted Input Parameters:

sigmas

This parameter represents the tensor of sigma values that you want to adjust. Sigma values are crucial in controlling the noise levels during the diffusion process in AI models. The input should be a tensor of floating-point numbers.

divide_by_last_sigma

This boolean parameter determines whether the adjusted sigma values should be normalized by the last sigma value in the tensor. If set to True, the adjusted sigma values will be divided by the last sigma value, ensuring that the last value is 1.0. This can help in maintaining a consistent scale for the sigma values. The default value is False.

divide_by

This parameter specifies a divisor by which all adjusted sigma values will be divided. It is a floating-point number that allows you to scale down the sigma values uniformly. The default value is 1.0, meaning no scaling will be applied unless specified otherwise.

offset_by

This integer parameter determines the offset index used to adjust the sigma values. Each sigma value will be replaced by the value at the index offset by this parameter. If the offset index is out of bounds, a small value (0.0001) will be used instead. This allows for more complex adjustments to the sigma values, potentially improving the noise schedule. The default value is 1.

Flip Sigmas Adjusted Output Parameters:

adjusted_sigmas

This output is the tensor of adjusted sigma values after applying the flipping, offsetting, and optional normalization. These values can be used directly in your AI model's noise scheduling process.

array_string

This output is a string representation of the adjusted sigma values, formatted for easy readability. It provides a quick way to inspect the adjusted values and ensure they meet your expectations.

Flip Sigmas Adjusted Usage Tips:

• To maintain a consistent scale for your sigma values, consider setting divide_by_last_sigma to True.
• Use the offset_by parameter to experiment with different noise schedules and observe how they affect the quality of your generated images.
• If you need to scale down the sigma values uniformly, adjust the divide_by parameter accordingly.

Flip Sigmas Adjusted Common Errors and Solutions:

"IndexError: index out of range"

• Explanation: This error occurs when the offset_by parameter results in an index that is out of the bounds of the sigma tensor.
• Solution: Ensure that the offset_by parameter is set to a value that keeps the index within the bounds of the sigma tensor.

"TypeError: expected Tensor as input"

• Explanation: This error occurs when the input sigmas is not a tensor.
• Solution: Make sure to provide a tensor of floating-point numbers as the sigmas input.

"ZeroDivisionError: division by zero"

• Explanation: This error occurs if divide_by is set to zero.
• Solution: Ensure that the divide_by parameter is set to a non-zero value to avoid division by zero errors.