ComfyUI Node: Guidance Limiter

Guidance Limiter:

The Guidance Limiter node is designed to enhance the quality of samples and distribution in diffusion models by applying guidance within a specified interval. This node leverages the method proposed by Kynkäänniemi et al., which suggests that limiting the application of guidance to certain intervals can significantly improve the results. By controlling the guidance application through parameters such as sigma_start, sigma_end, and cfg_rescale, you can fine-tune the model's behavior to achieve more desirable outcomes. This node is particularly useful for AI artists looking to refine their diffusion models to produce higher quality and more consistent results.

Guidance Limiter Input Parameters:

model

This parameter represents the diffusion model that you want to apply the guidance limiter to. It is essential as it forms the basis upon which the guidance adjustments will be made.

sigma_start

This parameter sets the starting point of the sigma interval where the guidance will be applied. If the sigma value is greater than sigma_start, the guidance scale will be set to 1. The default value is 5.42, with a minimum of -1.0 and a maximum of 10000.0. Adjusting this value allows you to control when the guidance begins to take effect.

sigma_end

This parameter sets the ending point of the sigma interval where the guidance will be applied. If the sigma value is less than or equal to sigma_end, the guidance scale will be set to 1. The default value is 0.28, with a minimum of -1.0 and a maximum of 10000.0. This parameter helps you define the range within which the guidance is active.

cfg_rescale

This parameter controls the rescaling of the classifier-free guidance (CFG). It ranges from 0.0 to 1.0, with a default value of 0.0. A higher value means more rescaling is applied, which can help in balancing the influence of the guidance. This parameter is crucial for fine-tuning the intensity of the guidance effect.

Guidance Limiter Output Parameters:

model

The output is the modified diffusion model with the guidance limiter applied. This model will have the guidance adjustments based on the specified sigma interval and rescaling parameters, leading to potentially improved sample quality and distribution.

Guidance Limiter Usage Tips:

• Experiment with different sigma_start and sigma_end values to find the optimal interval for your specific model and dataset. This can significantly impact the quality of the generated samples.
• Use the cfg_rescale parameter to fine-tune the intensity of the guidance. Start with the default value and gradually adjust it to see how it affects the results.
• Combine the Guidance Limiter with other model patches and samplers to explore various configurations and achieve the best possible outcomes for your AI art projects.

Guidance Limiter Common Errors and Solutions:

"Invalid sigma_start or sigma_end value"

• Explanation: This error occurs when the sigma_start or sigma_end values are set outside the acceptable range.
• Solution: Ensure that sigma_start and sigma_end are within the range of -1.0 to 10000.0. Adjust the values accordingly and try again.

"Model not provided"

• Explanation: This error occurs when the model parameter is not supplied.
• Solution: Make sure to provide a valid diffusion model as the input to the node. Check that the model is correctly loaded and passed to the node.

"CFG rescale value out of range"

• Explanation: This error occurs when the cfg_rescale value is set outside the range of 0.0 to 1.0.
• Solution: Adjust the cfg_rescale value to be within the specified range. The value should be between 0.0 and 1.0 to ensure proper functioning.