FunPack Hybrid Euler 2S Sampler

Sophisticated hybrid sampling node for quality motion dynamics with progressive correction blending and optional features for high-fidelity outputs.

FunPack Hybrid Euler 2S Sampler:

The FunPackHybridEuler2SSampler is a sophisticated hybrid sampling node designed to enhance the quality and motion dynamics of generated content. It combines an early Euler ancestral method with order-2 denoised extrapolation for motion, transitioning to a late DPM-Solver++(2S) ODE for quality refinement. This dual approach allows for progressive correction blending, ensuring high-quality outputs with smooth motion transitions. The sampler is equipped with optional features such as eta decay, anti-stiffness motion pulses, velocity bias, and reactive rescue mechanisms. These features make it particularly effective for applications requiring high fidelity and dynamic motion, such as video and audio processing. The node is also adaptable to rectified-flow models, where it employs a Heun corrector instead of the standard 2S method, ensuring compatibility and enhanced performance across different model types.

FunPack Hybrid Euler 2S Sampler Input Parameters:

eta

The eta parameter controls the noise level during the sampling process, influencing the randomness and diversity of the generated output. A higher eta value introduces more noise, potentially leading to more varied results, while a lower value results in more deterministic outputs. The default value is 1.0, with a typical range from 0.0 to 2.0.

s_noise

The s_noise parameter determines the strength of the noise applied during sampling. It affects the texture and detail of the output, with higher values leading to more pronounced noise effects. The default value is 1.0, and it can be adjusted to fine-tune the balance between noise and clarity.

high_quality_pct

This parameter specifies the percentage of the sampling process dedicated to high-quality refinement. A higher percentage allocates more steps to quality enhancement, improving the final output's sharpness and detail. The default value is 0.35, indicating that 35% of the process focuses on quality.

correction_blend

correction_blend controls the blending ratio between the initial and corrected outputs during the sampling process. A value of 1.0 applies full correction, while lower values blend the original and corrected outputs. The default is 1.0, ensuring maximum correction.

quality_sharpness

The quality_sharpness parameter adjusts the sharpness of the output during the quality refinement phase. Higher values increase sharpness, enhancing detail and contrast, while lower values produce softer results. The default is 0.0, providing a neutral sharpness level.

quality_sigma_start

This parameter sets the starting sigma value for the quality refinement phase, influencing the initial noise level. A lower sigma value results in less noise and a smoother start to the refinement process. The default is not specified, allowing for dynamic adjustment based on other parameters.

motion_pulse_mode

motion_pulse_mode determines the activation of motion pulses, which enhance dynamic motion effects. Options include "off" for no pulses and other modes for varying pulse intensities. The default is "off," disabling motion pulses.

motion_pulse_start_pct

This parameter defines the starting point of motion pulses as a percentage of the total sampling process. It determines when motion effects begin to influence the output. The default value is 0.30, starting pulses at 30% of the process.

motion_pulse_count

motion_pulse_count specifies the number of motion pulses applied during sampling. More pulses can enhance motion dynamics but may also increase processing time. The default is 2, providing a moderate level of motion enhancement.

motion_pulse_spacing_pct

This parameter sets the spacing between motion pulses as a percentage of the total process. It controls the frequency of motion effects, with smaller values leading to more frequent pulses. The default is 0.22, spacing pulses evenly throughout the process.

motion_pulse_strength

motion_pulse_strength determines the intensity of motion pulses, affecting the prominence of motion effects. Higher values result in stronger motion dynamics. The default is 0.85, providing a noticeable but balanced motion effect.

motion_pulse_noise

This parameter adds noise to motion pulses, enhancing their randomness and variability. A value of 0.0 applies no additional noise, while higher values increase variability. The default is 0.0, maintaining consistent motion effects.

motion_pulse_steps

motion_pulse_steps specifies the number of steps over which motion pulses are applied. More steps can smooth out motion effects but may also extend processing time. The default is not specified, allowing for dynamic adjustment.

velocity_bias_mode

This parameter controls the application of velocity bias, which influences the direction and speed of motion effects. Options include "off" for no bias and other modes for varying bias levels. The default is "off," disabling velocity bias.

velocity_bias_strength

velocity_bias_strength determines the intensity of velocity bias, affecting the directionality and speed of motion effects. Higher values increase bias strength. The default is 0.0, applying no bias.

velocity_bias_source

This parameter specifies the source of velocity bias, influencing how bias is calculated and applied. Options include "mean" for average-based bias and other sources for different calculations. The default is "mean," using average-based bias.

velocity_refinement_key

velocity_refinement_key sets the key for velocity refinement, determining how velocity effects are refined during sampling. The default is "default," applying standard refinement.

rescue_mode

This parameter activates rescue mode, which applies corrective measures to prevent undesirable artifacts. When enabled, it can improve output quality in challenging scenarios. The default is False, disabling rescue mode.

rescue_threshold

rescue_threshold sets the threshold for activating rescue mode, determining when corrective measures are applied. A lower threshold triggers rescue mode more frequently. The default is 0.15, providing a moderate activation level.

rescue_strength

This parameter determines the strength of rescue mode corrections, affecting the intensity of corrective measures. Higher values apply stronger corrections. The default is 0.2, providing a balanced correction level.

rescue_prompt_sig

rescue_prompt_sig specifies the sigma value for rescue mode prompts, influencing the noise level during corrections. The default is not specified, allowing for dynamic adjustment.

eta_final

The eta_final parameter sets the final noise level at the end of the sampling process, influencing the randomness of the final output. A higher value increases noise, while a lower value results in a more deterministic output. The default is 1.0.

normalize_strength

normalize_strength determines the strength of normalization applied during sampling, affecting the consistency and uniformity of the output. Higher values increase normalization strength. The default is 0.0, applying no normalization.

normalize_start_sigma

This parameter sets the starting sigma value for normalization, influencing the initial noise level during normalization. A lower sigma value results in less noise and a smoother start to normalization. The default is 0.9.