Nodes/MilitantHitchhiker-SwitchbladePack/GOD Scheduler (Advanced)
ComfyUI Node

GOD Scheduler (Advanced)

SIGMAS Straight From the Model, No Frills

By MilitantHitchhiker·Created 2 years ago·Updated 11 months ago· 3
GOD Scheduler (Advanced)
  • model
  • sigmas
steps28
denoise1.00

The GOD Scheduler (Advanced) (class GOD Scheduler (Advanced)) is the most boring node in this pack's generation corner, and that's not an insult. Its whole job is to hand you a descending SIGMAS schedule - high noise to low noise, computed from your loaded model's own sigma bounds - so you can drive an advanced sampler with a schedule you actually control. If you've ever wired a KSampler (Advanced) and wondered where its sigmas come from, this is a visible version of that machinery.

What it is and why you'd bother

The scheduler (noise schedule) decides how much noise gets removed at each step; Karras, linear, exponential and the rest are different recipes for that curve. This node is a recipe of its own: it takes the model's sigma_max/sigma_min, interpolates log-space between them, and guarantees a strictly descending run of steps values. denoise compresses the path toward the start when you want partial denoising (img2img-style). That's the entire feature set - no warp, no seed, no hidden state. Same model, steps, and denoise always yield the same sigmas.

So why reach for it instead of just picking "normal" in the KSampler dropdown? Two reasons. One: it feeds the advanced path - SamplerCustom and friends take explicit SIGMAS, and the stock KSampler (Advanced) wants a schedule too. Two: it's the intended pair for this pack's GOD samplers, which do the interesting work at sampling time and will consume whatever schedule you hand them.

Inputs and output

  • model - MODEL; only its sigma bounds are read, but it's required, so wire the model you're about to generate with.
  • steps - 28 default, range up to 2000.
  • denoise - 1.0 default; below that shortens the travelled noise path.

Output is sigmas (SIGMAS), which goes into KSampler (Advanced)'s sigmas input or any custom sampler that accepts a schedule.

Installing it

It ships in the Switchblade Pack. Via ComfyUI Manager, search MilitantHitchhiker-SwitchbladePack and install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MilitantHitchhiker/MilitantHitchhiker-SwitchbladePack

Restart ComfyUI. Nothing to download; it's pure torch math.

The honest take

Standalone, this is a mild variation on what KSampler already does internally - a log-space descending schedule is close to the "normal" you already have. The value is structural: it's the plumbing that lets the pack's GOD samplers plug into the advanced path, and it gives you a reproducible schedule object to swap between experiments. If you're not going to use the GOD or ODE samplers, you probably don't need this node.

One thing worth knowing before you judge results: the KB's sampler guidance says schedule shape matters most for SD 1.5/SDXL, while flow-matching models like Flux are picky about aggressive redistribution of denoising effort. This schedule is conservative by design, which is the right instinct. And a provenance note, because it matters for trust: this is a single-author experiment with zero community discussion behind it - nobody has written up 10,000 images of results with this scheduler. Use it as a control or a curiosity, A/B against your default, and don't let a mystery scheduler quietly become your default.

CategoryMilitantAI/Switchblade/Generation

Inputs (3)

NameTypeDefaultDescription
modelMODEL
stepsINT281–2000
denoiseFLOAT1.000–1

Outputs (1)

NameTypeDescription
sigmasSIGMAS