CustomSigmas Node Documentation

Overview

The CustomSigmas node is part of the ComfyUI-KJNodes collection, designed to facilitate advanced control over sigma schedules used in diffusion models. This node allows users to create a custom sigma schedule from a string of comma-separated values, offering more precise control over the diffusion process. This can be useful in various diffusion-based workflows, such as image generation or denoising tasks.

Functionality

The primary purpose of the CustomSigmas node is to enable the creation of a custom tensor of sigmas, which can then be used in conjunction with diffusion algorithms. Sigmas play a crucial role in these algorithms as they control the progression of noise reduction across steps. By customizing sigmas, users can potentially improve or tailor the behavior of their diffusion models.

Inputs

• Sigmas String: A multiline string comprising a sequence of comma-separated numeric values. Each value represents a sigma at a particular step. For example:

  • Nvidia's optimized AYS 10 step schedule for SD 1.5:

    14.615, 6.475, 3.861, 2.697, 1.886, 1.396, 0.963, 0.652, 0.399, 0.152, 0.029
    

  • SDXL:

    14.615, 6.315, 3.771, 2.181, 1.342, 0.862, 0.555, 0.380, 0.234, 0.113, 0.029
    

• Interpolate to Steps: An integer value representing the number of steps to interpolate the sigma values to. It helps in adjusting the granularity of the sigma sequence to fit the desired number of processing steps.

Outputs

• SIGMAS: The primary output is a tensor encapsulating the custom sigma values following interpolation. This tensor can be fed into other nodes or processes that utilize sigma schedules for diffusion-related tasks.

Use in ComfyUI Workflows

The CustomSigmas node can be integrated into various ComfyUI workflows, particularly those involving diffusion models. The node is especially beneficial when:

• There is a need to alter the noise reduction schedule to better fit specific dataset characteristics or application goals.
• Users want to experiment with different diffusion schedules to observe their impact on final outputs.
• Custom schedules are required to match or emulate the behavior of specific diffusion techniques or models, such as those proposed by Nvidia or in SDXL configurations.

Special Features and Considerations

• Customizability: The node offers significant flexibility to users by allowing custom sigma sequences. This feature is particularly useful for advanced users who want to tweak model behavior at a granular level.
• Interpolation Capability: It provides an easy mechanism to interpolate sigma values across a specified number of steps, aiding in better integration with existing workflows that may have a fixed number of diffusion steps.
• Optimization: By adjusting sigma values, users can optimize workflow performance and quality, potentially resulting in faster convergence or improved output quality.

In summary, the CustomSigmas node is a powerful tool for those looking to exert detailed control over the sigma schedules in diffusion-based workflows within the ComfyUI environment. It combines ease of use with advanced customization, making it an invaluable component for both beginners and seasoned users.