Gear Sigmas (explicit list)

Gear Sigmas is the laziest node in the Gear pack, and that's the point. One text box, one output. You type a comma-separated list of sigmas and it hands you a SIGMAS tensor you can plug into any sampler that takes one. No scheduler math, no hidden work, nothing to tune. The name oversells it - it's a clipboard with a type check.

So why would you reach for a node this boring? Because distilled video models like the LTX family run on a fixed, recipe-validated schedule, and the stock scheduler nodes don't always give you that schedule. A scheduler computes sigmas from a step count and a curve; a recipe author has usually already done the validation and hands you the exact numbers. And as the whole sampler-tuning conversation keeps finding, aggressive schedulers actively hurt on distilled flow-matching models - the straighter the denoising trajectory, the more a reshaped schedule costs you. When someone says "use this exact list, it's what was tested," you want a node that lets you obey.

How it works

It's plumbing. The node takes the string, strips whitespace, splits on commas, parses each value to a float, and returns a SIGMAS tensor. The author's contract, straight from the tooltip: comma-separated, descending, ending in 0. If the list doesn't end in 0 - or has fewer than two values - it raises Gear Sigmas: need a descending list ending in 0. so you find out immediately instead of sampling nonsense silently.

The default that ships with the node is the pack's validated 8-step distilled schedule, which the README labels the LTX 2.5 distilled schedule:

1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0

Count the values: nine numbers, eight steps. That's the whole game for LTX distilled workflows - 8 steps, CFG 1, an explicit schedule.

The one input and the one output

• sigmas - a STRING (single-line). Paste the schedule. That's it.
• SIGMAS - the output. Wire it into Gear Tiled Fusion Sampler (its natural partner in this pack), or any node with a SIGMAS socket like SamplerCustom / SamplerCustomAdvanced. It's the same type a KSampler's scheduler produces internally, so nothing downstream cares where it came from.

Installing it

It's part of the ComfyUI_Gear pack, so you're installing the whole thing:

• ComfyUI Manager: search for ComfyUI_Gear and install, then restart.
• Manual:

  cd ComfyUI/custom_nodes
  git clone https://github.com/oumad/ComfyUI_Gear
  cd ComfyUI_Gear
  pip install -r requirements.txt

  Then restart ComfyUI. Dependencies are light (numpy, Pillow, OpenEXR, opencv) - nothing model-sized to download for this node.

Gotchas

• The list must end in 0. Paste a schedule that trails off at 0.421875 and the node throws. It's deliberate - the final sigma has to be zero or the sampler never finishes the job.
• Format is strict but forgiving. Spaces are stripped, so 1.0, 0.9, 0.0 and 1.0,0.9,0.0 both work. Brackets, semicolons, or scientific-notation weirdness will not.
• This is a distilled-schedule node. For a non-distilled dev model you'd use a normal scheduler at 20+ steps; pasting a distilled schedule onto a dev model is asking for a mess.

If you just want to paste the schedule your workflow's recipe demands and move on, this is the least-friction way to do it in ComfyUI.