🔥 Winnougan LTX Sigma Schedule

There's a moment in every LTX-2.3 workflow where you're staring at a ManualSigmas node with a string like 1.0, 0.99375, 0.9875, 0.98125, 0.975, 0.909375, 0.725, 0.421875, 0.0 and wondering where it came from. That's what the Winnougan LTX Sigma Schedule node eliminates. Pick a preset from a dropdown, wire the output to SamplerCustomAdvanced, and the correct sigma tensor - plus the matching step count - is handled for you.

Why this exists

LTX-2.3's fast path is the distilled model, which per the model card runs at roughly 8 steps with CFG 1. That's the whole deal: distilling collapsed the 20–40 step full-model trajectory down to a handful of big jumps, and the jump points are non-negotiable - the schedule is a specific list of sigma values, not a generic basic scheduler. The pack author pulled these from Lightricks' official LTX-2.3 workflows, so the presets are the ones the model was actually tuned around, and the two-stage design (Stage 1 main generation, Stage 2 after the spatial upscaler) is baked into the preset names.

Inputs and outputs

• preset - four real presets plus Custom:
  • Stage 1 - Distilled 4-step (fast): 0.85, 0.725, 0.421875, 0.0. The default, and the one you'll use for most main generations.
  • Stage 1 - Distilled 8-step (quality): the nine-value list above. More detail, slower.
  • Stage 2 - Upscaler 3-step and Upscaler 4-step: low-denoise refinement passes to run after a spatial upscaler, not for main generation.
• custom_sigmas - comma-separated values, only read when preset is Custom. The default is pre-filled with the 4-step Stage 1 list so you can see the format.

Outputs are sigmas (a SIGMAS tensor → SamplerCustomAdvanced), steps (an int → your sampler's step count), and schedule_info (a string → Show Text, handy for pasting into workflow notes).

Wiring it

Stage 1 and Stage 2 are separate nodes in practice. The main sampler takes the Stage 1 sigmas and steps; after you upscale, a second SamplerCustomAdvanced takes the Stage 2 output. If you wire a Stage 2 schedule into the main pass you'll get a mostly-deep-denoise generation that looks wrong in a way that's hard to diagnose. Wire steps to the sampler too - a mismatch between the sigma count and the step count is the classic error here, and the node hands you both so you can't get it wrong.

Install

This is one of eight nodes in the ComfyUI_WLTX_nodes pack, all under the Winnougan LTX category:

cd ComfyUI/custom_nodes
git clone https://github.com/Winnougan/ComfyUI_WLTX_nodes

Restart ComfyUI after cloning, or search ComfyUI_WLTX_nodes in ComfyUI Manager. No Python dependencies beyond what ComfyUI already ships - the node is pure torch, producing the tensor from a hardcoded preset list.

Common issues

If you're coming from the official LTX workflows, note this replaces a ManualSigmas node - you still need SamplerCustomAdvanced (or the equivalent advanced sampler), not the plain KSampler. And when you're tuning by hand, use Custom rather than mangling a preset: one bad value in a comma-separated list throws a clear "Invalid sigma value" error, which is the node doing you a favor before you burn a generation on it.