Step Split Override
Take exact control of the high/low handoff
- plan
- plan
At some point in a Wan 2.2 workflow you stop trusting the profile's default split and want to say "no, the high-noise expert runs exactly N steps." Step Split Override is the node for that. It's the successor to Range Split Override (Legacy) and it replaces that node's fuzzy percentage with the thing you actually mean: an exact count of sampling transitions assigned to the high-noise expert.
What it does
Inputs: plan and steps_high (default 0, 0–9999). The 0 default is the clever part - it means pass the plan through unchanged. That's the recommended way to leave the node visibly in-chain without applying a manual split, so you can wire it up now and start forcing values later without re-plumbing the graph.
Set steps_high to a positive value and the node rebuilds the complete plan: the whole curve, both sigma slices, and the step split, all revalidated. It replaces Priority's calculated high-stage transition count with yours.
The acceleration nuance
How the remaining steps shake out depends on which stages are accelerated, and this is where people get surprised. From the README:
- Uniform budgets (
High + LoworNone) - both stages draw from one budget, so your forced value moves the split point while preserving the plan's effective total. - Mixed acceleration (
High only/Low only) - the stages draw from different budgets, so the forced value overrides the high allocation only, and the low stage keeps its own budget share. The example given:Low onlywith budgets A10/F30 plans 14 high + 5 low; forcingsteps_high = 5yields 5 high + 5 low - the remainder is not redistributed into the accelerated CFG-1 low stage. The forced value may use up to the full high-stage budget, and that's it.
So in mixed mode, think of this as "I want the high stage to do exactly this many steps" rather than "I want to rebalance everything."
Ordering and interaction
It's order-independent with Shift Override - each stores its request and rebuilds from the original planner controls, so either order gives the same result. With Progressive 50/50 Sigma Control the README recommends a specific order: the progressive override feeds the plan's sigma_override first, then Step Split Override, then Sigma Breakout. In that arrangement, steps_high = 0 keeps the accelerated 50/50 sigma plan, and any positive value manually overrides the split within that accelerated total.
Install
Pack-standard - 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.
The verdict: this is the split-control node you actually want. The 0 = passthrough behavior means there's zero risk in wiring it in preemptively, and when you do want to pin the handoff, you get an exact number instead of a percentage that only approximately maps to steps.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| plan | WAN22_SAMPLING_PLAN | The Wan 2.2 plan to refine. | |
| steps_high | INT | 00–9999 | Exact number of sampling transitions assigned to the high-noise expert. Use 0 to pass the plan through unchanged. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| plan | WAN22_SAMPLING_PLAN | — |