Node for Blending Sigmas

[recris]

Feature Idea

Different schedulers have different strengths and weaknesses. In my experiments with photography/realism prompts using Flux.1 Dev I've noticed that beta scheduler has very good prompt adherence but the output tends to be less diverse (certain prompts have a very similar look and feel across different seeds). By contrast ddim_uniform is a bit worse when it comes to adherence but the output looks more varied and creative and photographic images also look more "natural".

The question is: can we combine the best of both schedules? I think having a way to blend the schedule sigmas using linear interpolation can work for this purpose (to some extent).

Existing Solutions

I've hacked together a modification of BasicScheduler node that implements the requested effect:

class BlendedScheduler:
    @classmethod
    def INPUT_TYPES(s):
        return {"required":
                    {"model": ("MODEL",),
                     "scheduler1": (comfy.samplers.SCHEDULER_NAMES, ),
                     "scheduler2": (comfy.samplers.SCHEDULER_NAMES, ),
                     "ratio": ("FLOAT", {"default": 0.5, "min": 0.0, "max": 1.0, "step": 0.01}),
                     "steps": ("INT", {"default": 20, "min": 1, "max": 10000}),
                     "denoise": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
                      }
               }
    RETURN_TYPES = ("SIGMAS",)
    CATEGORY = "sampling/custom_sampling/schedulers"

    FUNCTION = "get_sigmas"

    def get_sigmas(self, model, scheduler1, scheduler2, ratio, steps, denoise):
        total_steps = steps
        if denoise < 1.0:
            if denoise <= 0.0:
                return (torch.FloatTensor([]),)
            total_steps = int(steps/denoise)

        sigmas1 = comfy.samplers.calculate_sigmas(model.get_model_object("model_sampling"), scheduler1, total_steps).cpu()
        sigmas1 = sigmas1[-(steps + 1):]
        sigmas2 = comfy.samplers.calculate_sigmas(model.get_model_object("model_sampling"), scheduler2, total_steps).cpu()
        sigmas2 = sigmas2[-(steps + 1):]
        
        sigmas = torch.lerp(sigmas1, sigmas2, ratio)
        return (sigmas, )

It probably makes more sense to have something that takes the sigmas as input instead?

Other

No response