IterativeMixingScheduler

This node is gloriously unglamorous, and that's fine. IterativeMixingSampler produces a SAMPLER object, but SamplerCustom needs both a sampler and a set of SIGMAS - the schedule of noise levels to walk down. This node is the pack's bundled way to generate those sigmas. It has no iterative-mixing logic of its own; it's a thin, correct wrapper around ComfyUI's standard sigma calculation. Its only job is to feed the mixer.

Inputs and output

• model - required, and not just for show. The sigma schedule is computed from the model's own noise sampling setup, so this has to be the same model you're sampling with.
• scheduler - the usual six: normal, karras, exponential, sgm_uniform, simple, ddim_uniform. normal is the default pick that matches the model's training; karras is the community default for most SD models if you want marginally snappier steps.
• steps (default 20) - total de-noising steps. The README's general tip for iterative mixing is 20–40 steps; fewer gives "interesting" but increasingly unhinged output.
• denoise (default 1.0) - the amount of the schedule to actually run. Below 1.0 it recalculates over a longer base schedule and keeps the tail, same as a core KSampler's denoise.

Output: SIGMAS, which you wire into the sigmas input of SamplerCustom (and typically into a VAEDecode-less path - the latent comes out of SamplerCustom).

Why it matters for iterative mixing

The sampler fits its blending curve to the sigmas you hand it. Feed it a full 40-step schedule and alpha_1 behaves one way; feed it a half schedule and the curve stretches. That's exactly why the pack also ships IterativeMixingSchedulerAdvanced - if you're only sampling part of the schedule, you need to think about the ratio (the advanced node's README note walks through the arithmetic). For a plain full-schedule run, this simple version is all you need and there's nothing to misconfigure.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/ttulttul/ComfyUI-Iterative-Mixer
# restart ComfyUI

Or via ComfyUI Manager as "ComfyUI Iterative Mixing Nodes." No model downloads; the pack's real dependencies (matplotlib, scipy) aren't even touched by this node, but requirements.txt installs them anyway.

One gotcha worth knowing: because this is a custom-sampler scheduler, nothing applies the denoise-style math for you downstream - the sampler runs exactly the steps in the sigmas. If you set denoise to 0.5 here, that's the whole story, no hidden magic. Keep add_noise off on SamplerCustom (the iterative mixing sampler wants to inject its own noise) and you're set.