ComfyUI Node: MiniMax H3 Dynamic Guidance / 动态引导 (Advanced)

MiniMax H3 Dynamic Guidance / 动态引导 (Advanced):

The MiniMaxH3DynamicCFGGuiderT8Advanced node is designed to provide dynamic guidance in audio processing workflows, specifically tailored for advanced users seeking to optimize their creative outputs. This node acts as a default-pass-through guider, applying a single device-side sigma-dependent gain to the BasicGuider route. It is not a true Classifier-Free Guidance (CFG) but rather a sophisticated method that requires a layout-matched negative and explicit cost consent. The node is experimental and aims to enhance the quality of motion and detail in audio outputs by dynamically adjusting guidance parameters based on the input conditions. Its primary goal is to offer a flexible and powerful tool for artists to refine their audio projects with precision and control.

MiniMax H3 Dynamic Guidance / 动态引导 (Advanced) Input Parameters:

mode

The mode parameter determines the operational mode of the node. It is set to "passthrough_basic" by default, which means the node will apply basic guidance without additional modifications. This parameter is crucial for defining how the node interacts with the input data and influences the overall guidance process.

early_scale

The early_scale parameter is a numerical value that adjusts the gain applied during the early stages of the guidance process. It has a default value of 1.0, indicating no additional scaling. This parameter allows users to fine-tune the initial impact of the guidance, potentially enhancing the early phases of audio processing.

late_scale

Similar to early_scale, the late_scale parameter adjusts the gain applied during the later stages of the guidance process. With a default value of 1.0, it ensures consistent scaling throughout the process. Adjusting this parameter can help refine the final output by modifying the influence of guidance in the concluding stages.

accept_true_cfg_cost

This boolean parameter, set to False by default, determines whether the node should accept the cost associated with true CFG. By default, the node does not engage in true CFG, but this parameter allows users to opt-in if they wish to explore this advanced feature.

accept_turbo_guidance_ood

The accept_turbo_guidance_ood parameter is a boolean flag that, when set to False, indicates that the node will not accept out-of-distribution (OOD) turbo guidance. This parameter is essential for users who want to maintain control over the guidance process and avoid unexpected results from OOD inputs.

MiniMax H3 Dynamic Guidance / 动态引导 (Advanced) Output Parameters:

av_latent

The av_latent output parameter represents the processed audio-visual latent data. This output is crucial for subsequent nodes in the workflow, as it carries the refined guidance information that has been applied to the input data. Understanding this output helps users track the changes made by the node and assess the impact on the overall project.

report_json

The report_json output provides a detailed JSON report of the guidance process. This report includes information about the guider calls, physical model forwards, and branch batching. It is an invaluable resource for users who wish to audit the guidance process and understand the internal workings of the node.

MiniMax H3 Dynamic Guidance / 动态引导 (Advanced) Usage Tips:

• Experiment with the early_scale and late_scale parameters to find the optimal balance for your specific audio project. Adjusting these values can significantly impact the quality and detail of the output.
• Use the accept_true_cfg_cost parameter cautiously. While it offers advanced CFG capabilities, it may introduce additional complexity and cost to the guidance process.

MiniMax H3 Dynamic Guidance / 动态引导 (Advanced) Common Errors and Solutions:

"Invalid mode selected"

• Explanation: This error occurs when an unsupported mode is set for the mode parameter.
• Solution: Ensure that the mode parameter is set to "passthrough_basic" or another valid option supported by the node.

"Scale value out of range"

• Explanation: This error indicates that the early_scale or late_scale parameter is set to a value outside the acceptable range.
• Solution: Verify that both early_scale and late_scale are set to values within the expected range, typically around 1.0 for default behavior.

"True CFG cost not accepted"

• Explanation: This error arises when the node attempts to apply true CFG without the accept_true_cfg_cost parameter being enabled.
• Solution: If you wish to use true CFG, set the accept_true_cfg_cost parameter to True. Otherwise, ensure that the node is configured to operate without true CFG.