KJNodes/noisegenerated

Custom Sigmas(CustomSigmas)

This node takes a comma-separated string of numerical values and converts it into a PyTorch tensor.

Custom Sigmas

SIGMAS
sigmas_string
14.615, 6.475, 3.861, 2.697, 1.886, 1.396, 0.963, 0.652, 0.399, 0.152, 0.029
interpolate_to_steps
10
KJNodes

The Custom Sigmas node receives a string of comma-separated numerical values and converts it into a PyTorch tensor.

The Custom Sigmas node allows users to create a noise schedule (sigmas) tensor by inputting a custom string of numerical values. It is primarily used to precisely control the noise intensity at each step during the sampling process of diffusion models (such as Stable Diffusion). This node provides the ability to fine-tune the generation process.

Node Function

This node receives a string of comma-separated numerical values and converts it into a PyTorch tensor. If the number of provided values does not match the target number of steps, the node automatically adjusts the sequence length using a log-linear interpolation algorithm to ensure the final output noise intensity sequence is smooth and matches the expected number of steps. This enables it to adapt to the preset optimized scheduling schemes of different models (e.g., SD 1.5, SDXL, SVD).

Node Parameter Description - Custom Sigmas

Control Parameters (Parameters)

Parameter NameData TypeRequiredDefault ValueRange/OptionsDescription
sigmas_stringSTRINGYes14.615, 6.475, 3.861, 2.697, 1.886, 1.396, 0.963, 0.652, 0.399, 0.152, 0.029-sigmas_string: Input a custom noise intensity sequence in the format of a comma-separated string of numerical values. There is no range restriction, but the values are typically positive and decreasing.
interpolate_to_stepsINTYes100 - 255 (Step: 1)interpolate_to_steps: Specifies the target number of sampling steps. The node will interpolate or adjust the input noise sequence accordingly. The range is from 0 to 255, with a step size of 1.

Output

Parameter NameData TypeDescription
SIGMASSIGMASSIGMAS: Outputs the adjusted noise schedule tensor, which can be directly used for noise control in the sampler.

Usage Scenarios

In workflows that require precise control over the image generation process or experimentation with different noise schedules, you can connect this node to the "sigmas" input port of a KSampler or KSampler Advanced. For example, you can directly input the 10-step schedule values optimized by Nvidia for SD 1.5 to attempt achieving better image quality with fewer sampling steps.

The Custom Sigmas node is from the ComfyUI-KJNodes node package.

Comments

Sign in with GitHub to join the discussion.

Loading comments…