ComfyUI Node: Custom Sigmas

Custom Sigmas:

The CustomSigmas node is designed to create a tensor of sigma values from a string of comma-separated values. This node is particularly useful for AI artists who need to generate custom noise schedules for their models. By providing a flexible way to adjust and interpolate sigma values, it allows for fine-tuning the noise levels used in various stages of the model's processing. This can be especially beneficial for optimizing the performance of models like Stable Diffusion (SD) and other similar architectures. The node also offers options to adjust the sigma values by dividing them, offsetting them, and flipping their order, providing a high degree of control over the noise schedule.

Custom Sigmas Input Parameters:

sigmas_string

This parameter takes a string of comma-separated values representing the sigma values. These values are used to create the initial tensor of sigmas. The default value is "14.615, 6.475, 3.861, 2.697, 1.886, 1.396, 0.963, 0.652, 0.399, 0.152, 0.029". This string should be formatted correctly to ensure proper parsing and conversion into a tensor.

interpolate_to_steps

This integer parameter specifies the number of steps to which the sigma values should be interpolated. The default value is 10, with a minimum of 0 and a maximum of 255. This parameter helps in adjusting the length of the sigma tensor to match the required number of steps for the model's processing.

divide_by_last_sigma

This boolean parameter determines whether the sigma values should be divided by the last sigma value in the tensor. The default value is False. When set to True, it normalizes the sigma values by the last value, which can be useful for certain types of noise schedules.

divide_by

This float parameter specifies the value by which all sigma values should be divided. The default value is 1, with a minimum of 1 and a maximum of 255. This parameter allows for scaling down the sigma values, which can be useful for fine-tuning the noise levels.

offset_by

This integer parameter determines the offset to be applied to the sigma values. The default value is 1, with a minimum of -100 and a maximum of 100. This parameter shifts the sigma values by the specified offset, allowing for more control over the noise schedule.

Custom Sigmas Output Parameters:

SIGMAS

This output parameter is a tensor of sigma values that have been adjusted based on the input parameters. The tensor is used in the model's processing to control the noise levels at different stages. The adjusted sigma values can significantly impact the model's performance and the quality of the generated outputs.

sigmas_string

This output parameter is a string representation of the adjusted sigma values. It provides a convenient way to visualize and verify the sigma values after they have been processed. The string format makes it easy to copy and reuse the values in other contexts or for further adjustments.

Custom Sigmas Usage Tips:

• Ensure that the sigmas_string is correctly formatted with comma-separated values to avoid parsing errors.
• Use the interpolate_to_steps parameter to match the sigma tensor length with the required number of steps for your model.
• Experiment with the divide_by_last_sigma and divide_by parameters to normalize and scale the sigma values for optimal performance.
• Adjust the offset_by parameter to shift the sigma values and explore different noise schedules.

Custom Sigmas Common Errors and Solutions:

"Invalid sigma string format"

• Explanation: The sigmas_string parameter is not correctly formatted.
• Solution: Ensure that the sigma values are comma-separated and correctly formatted as a string.

"Interpolation steps out of range"

• Explanation: The interpolate_to_steps parameter is set to a value outside the allowed range.
• Solution: Set the interpolate_to_steps parameter to a value between 0 and 255.

"Division by zero error"

• Explanation: The divide_by parameter is set to 0, causing a division by zero.
• Solution: Ensure that the divide_by parameter is set to a value greater than 0.

"Offset out of range"

• Explanation: The offset_by parameter is set to a value outside the allowed range.
• Solution: Set the offset_by parameter to a value between -100 and 100.