Sigma Schedule Interpolated Mean 🎭🅐🅓

Documentation

Class name: ADE_SigmaScheduleWeightedAverageInterp
Category: Animate Diff 🎭🅐🅓/sample settings/sigma schedule
Output node: False

This node is designed to create a new sigma schedule by interpolating between two given sigma schedules based on a specified weight range and interpolation method. It effectively blends the characteristics of both input schedules into a single, new schedule, allowing for dynamic adjustments in the diffusion process.

Input types

Required

schedule_A
- The first sigma schedule to be interpolated.
- Comfy dtype: SIGMA_SCHEDULE
- Python dtype: SigmaSchedule
schedule_B
- The second sigma schedule to be interpolated.
- Comfy dtype: SIGMA_SCHEDULE
- Python dtype: SigmaSchedule
weight_A_Start
- The starting weight for the first sigma schedule in the interpolation.
- Comfy dtype: FLOAT
- Python dtype: float
weight_A_End
- The ending weight for the first sigma schedule in the interpolation.
- Comfy dtype: FLOAT
- Python dtype: float
interpolation
- The method of interpolation to be used for blending the sigma schedules.
- Comfy dtype: COMBO[STRING]
- Python dtype: InterpolationMethod

Output types

sigma_schedule
- Comfy dtype: SIGMA_SCHEDULE
- The resulting sigma schedule after interpolation.
- Python dtype: SigmaSchedule

Usage tips

Infra type: CPU
Common nodes: unknown

Source code

class InterpolatedWeightedAverageSigmaScheduleNode:
    @classmethod
    def INPUT_TYPES(s):
        return {
            "required": {
                "schedule_A": ("SIGMA_SCHEDULE",),
                "schedule_B": ("SIGMA_SCHEDULE",),
                "weight_A_Start": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.001}),
                "weight_A_End": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.001}),
                "interpolation": (InterpolationMethod._LIST,),
            }
        }
    
    RETURN_TYPES = ("SIGMA_SCHEDULE",)
    CATEGORY = "Animate Diff 🎭🅐🅓/sample settings/sigma schedule"
    FUNCTION = "get_sigma_schedule"

    def get_sigma_schedule(self, schedule_A: SigmaSchedule, schedule_B: SigmaSchedule,
                           weight_A_Start: float, weight_A_End: float, interpolation: str):
        validate_sigma_schedule_compatibility(schedule_A, schedule_B)
        # get reverse weights, since sigmas are currently reversed
        weights = InterpolationMethod.get_weights(num_from=weight_A_Start, num_to=weight_A_End,
                                                  length=schedule_A.total_sigmas(), method=interpolation, reverse=True)
        weights = weights.to(schedule_A.model_sampling.sigmas.dtype).to(schedule_A.model_sampling.sigmas.device)
        new_sigmas = schedule_A.model_sampling.sigmas * weights + schedule_B.model_sampling.sigmas * (1.0-weights)
        combo_schedule = schedule_A.clone()
        combo_schedule.model_sampling.set_sigmas(new_sigmas)
        return (combo_schedule,)

3.1 KiB Raw Permalink Blame History