ComfyUI Node: KSampler (Spectrum + Mod Guidance Advanced)

KSampler (Spectrum + Mod Guidance Advanced):

SpectrumKSamplerAdvanced is a sophisticated node designed to enhance the sampling process by integrating modulation guidance and customizable forecasting parameters. This node is particularly beneficial for AI artists seeking to optimize the quality and speed of their image generation tasks. It leverages a spectrum-accelerated approach, which allows for skipping certain computational blocks, thereby achieving a significant speedup without compromising on quality. The node provides a comprehensive set of controls, including modulation guidance towards quality tags and adaptive SMC-CFG for detail recovery. It subsumes previous iterations of SpectrumKSampler nodes, offering a unified and advanced solution for sampling needs. The node is equipped with features like Foresight Guidance (FSG) and Chebyshev polynomial feature forecasting, which together enable a fine-tuned balance between speed and quality. This makes SpectrumKSamplerAdvanced an essential tool for those looking to push the boundaries of AI-generated art.

KSampler (Spectrum + Mod Guidance Advanced) Input Parameters:

adaptive_smc_alpha

This parameter controls the adaptive SMC-CFG, which is responsible for detail recovery during the sampling process. It adjusts the alpha value used in the sliding-mode CFG combine, allowing for more nuanced control over the detail level in the generated images. The default value is set to _SMC_CFG_ALPHA_DEFAULT, and it can be tuned to achieve the desired level of detail.

fsg

The fsg parameter is a boolean toggle that enables Foresight Guidance, a feature that guides the sampling process towards a predefined "golden path" for optimal results. When enabled, it activates a validated production stack with specific settings like CFG++ λ=1.5 and FSG band [0.59,0.75]. This feature is mutually exclusive with SMC-CFG, meaning it cannot be used simultaneously with it. The default value is False.

refresh_ratio

This parameter determines the Spectral-Evolution-Aware (SEA) skip decision, which influences the speed of the sampling process. A value of -1 turns off SEA, reverting to a plain growing window approach. Adjusting the refresh_ratio can lead to a 2-3x speedup by skipping unnecessary computations, making it a crucial parameter for optimizing performance.

mod_w_profile

mod_w_profile is used for optional per-block modulation guidance towards quality tags. When set to 'off', no steering is applied. This parameter allows users to influence the sampling process by guiding it towards specific quality attributes, enhancing the final output's aesthetic appeal.

quality_neg

This parameter is used to specify negative quality tags, which the modulation guidance will steer away from during the sampling process. It helps in avoiding certain undesirable features or qualities in the generated images, ensuring that the output aligns more closely with the user's artistic vision.

clip

The clip parameter is an optional input that can be used to provide additional context or guidance to the sampling process. It allows for more precise control over the generated output by incorporating external information or constraints.

denoise

denoise controls the level of noise reduction applied during the sampling process. A value of 1.0 applies full denoising, while lower values retain more noise, which can be useful for achieving certain artistic effects. This parameter is crucial for balancing clarity and texture in the final output.

KSampler (Spectrum + Mod Guidance Advanced) Output Parameters:

LATENT

The output of the SpectrumKSamplerAdvanced node is a LATENT parameter, which represents the latent space representation of the generated image. This output is crucial as it forms the basis for further processing or conversion into a final image. The latent representation encapsulates the essential features and qualities of the generated content, allowing for subsequent refinement or transformation.

KSampler (Spectrum + Mod Guidance Advanced) Usage Tips:

• Experiment with the adaptive_smc_alpha parameter to find the optimal balance between detail and speed for your specific project needs.
• Utilize the fsg toggle to enable Foresight Guidance when aiming for high-quality outputs with predefined settings, but remember to disable SMC-CFG as they cannot be used together.
• Adjust the refresh_ratio to maximize speed without sacrificing quality, especially when working with large datasets or complex models.
• Use mod_w_profile to guide the sampling process towards desired quality attributes, enhancing the aesthetic appeal of the final output.

KSampler (Spectrum + Mod Guidance Advanced) Common Errors and Solutions:

"Invalid CFG++ λ value"

• Explanation: This error occurs when the CFG++ λ value is set incorrectly, possibly due to a conflict with other parameters like fsg.
• Solution: Ensure that the fsg toggle is set correctly and that CFG++ λ is compatible with other settings. Adjust the λ value as needed to resolve the conflict.

"SMC-CFG and FSG cannot be used together"

• Explanation: This error indicates that both SMC-CFG and FSG are enabled simultaneously, which is not allowed.
• Solution: Disable either SMC-CFG or FSG to resolve the conflict. These features are mutually exclusive and cannot be used together.

"Invalid refresh_ratio value"

• Explanation: This error arises when the refresh_ratio is set to an unsupported value, affecting the SEA skip decision.
• Solution: Set the refresh_ratio to a valid value, ensuring it aligns with the desired speed and quality tradeoff. Avoid using -1 unless reverting to a plain growing window approach.