Nodes/Sampling Planner/Sigma Breakout
ComfyUI Node

Sigma Breakout

The authoritative sigmas when KSampler can't represent your plan

By boobkake22·Created 2 months ago·Updated 2 months ago· 0
Sigma Breakout
  • plan
  • sigmas_high
  • sigmas_low
  • sigmas_all
  • steps
  • steps_high
  • steps_low
  • shift
  • plan

Every other breakout in the Sampling Planner pack eventually answers to this one. Sigma Breakout is the authoritative way to get the actual sigma schedules out of a Wan 2.2 sampling plan - the exact validated curve, including the mixed-budget and priority-adjusted piecewise curves that ordinary KSampler controls simply cannot express. When KSampler Breakout tells you a plan isn't representable, this is where it points you.

What you get

One input: plan. Seven outputs:

  • sigmas_high - the high-noise expert's slice of the schedule
  • sigmas_low - the low-noise expert's slice (the high schedule ends at the same sigma the low schedule begins - that shared handoff is deliberate)
  • sigmas_all - the complete schedule
  • steps / steps_high / steps_low - the counts, handy for display or logging
  • shift - the plan's shift
  • plan - pass-through

The high and low slices feed SIGMAS inputs directly, which is exactly what the custom-sampler path needs. No widgets to mis-set; it's a pure breakout.

The workflow it enables

For two SamplerCustomAdvanced stages - the route the pack recommends when native KSampler controls can't reproduce the curve - the wiring is:

RandomNoise → high SamplerCustomAdvanced → output → low SamplerCustomAdvanced → output
                                                   ↑
                                              DisableNoise

Use the high sampler's output (not denoised_output) as the low stage's latent. This trips people up constantly: denoised_output is an x0 prediction meant for preview or inspection, not the partially-sampled latent at the shared handoff sigma. For KSampler parity, the pack's notes recommend a CFGGuider per stage with the same conditioning and CFG values your KSampler Advanced nodes used, the same sampler via KSamplerSelect, and - critically - no separate BasicScheduler between the plan and the samplers. Sigma Breakout already emits the slices; inserting another scheduler node is how you get double-scheduled curves.

When to reach for it

Two cases. First, KSampler Breakout refuses: piecewise plans, or a sampler that changes the core sigma count. Second, you're running a custom-sampler graph anyway and want the plan's exact curve rather than letting a scheduler rebuild something approximate. In both cases Sigma Breakout is the ground truth - it preserves the validated curve rather than approximating it, which is the whole reason it exists as the pack's "authoritative sigma-output path."

Install

Standard pack install - ComfyUI Manager (search Sampling Planner) or:

cd ComfyUI/custom_nodes
git clone https://github.com/boobkake22/ComfyUI-SamplingPlanner

restart; it's under sampling / Sampling Planner / Wan 2.2. No dependencies, no models.

One honest caveat: this path is more plumbing than the KSampler route - guiders, sampler nodes, noise nodes, all the pieces KSampler Advanced used to hide. If your plan is exactly representable, KSampler Breakout gives you the same curve with a fraction of the wiring. But when the plan gets fancy, this is the node that keeps the curve honest instead of silently approximating it.

Categorysampling/Sampling Planner/Wan 2.2

Inputs (1)

NameTypeDefaultDescription
planWAN22_SAMPLING_PLAN

Outputs (8)

NameTypeDescription
sigmas_highSIGMAS
sigmas_lowSIGMAS
sigmas_allSIGMAS
stepsINT
steps_highINT
steps_lowINT
shiftFLOAT
planWAN22_SAMPLING_PLAN