Step Budget

Step Budget is the smallest "real" node in the Sampling Planner pack: two integers in, one value out, and yet every acceleration decision your Wan 2.2 workflow makes flows from it. It's the node that answers the question "how many steps am I actually paying for?"

The two numbers

• accelerated_steps (default 10, 2–99) - the step count you'd use if the entire denoising range were accelerated
• full_steps (default 30, 2–99) - the step count you'd use if the entire range were unaccelerated

The output is a SAMPLING_STEP_BUDGET object that feeds Sampling Plan (Wan 2.2)'s step_budget input. That's the only thing it wires into, and it's all it does.

Why "full-range-equivalent" matters

Wan 2.2 splits denoising between a high-noise expert and a low-noise expert, and you might accelerate only one of them. So a single "steps" number is meaningless - what you actually need is a budget per stage: if this whole range were accelerated, how many steps? and if it weren't, how many? The planner then projects the right budget onto each expert's share of the range. With 10 accelerated / 30 full and a 50/50 split, you get:

| Acceleration | High stage | Low stage | Effective total | |---|---|---|---| | High + Low | 5 accelerated | 5 accelerated | 10 | | High only | 5 accelerated | 15 full | 20 | | Low only | 15 full | 5 accelerated | 20 | | None | 15 full | 15 full | 30 |

Same two inputs, four very different effective costs - which is exactly why the budget is expressed as range-equivalents rather than a single step count. A 10/30 budget is also the pack's documented baseline for the LightX2V-style recipe, so those defaults are a reasonable place to start.

Install

It's a helper, so it lives under sampling / Sampling Planner / Helpers. Pack-standard install:

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

or ComfyUI Manager (search Sampling Planner), then restart. No dependencies or models.

The honest framing: this node is just a tidy wrapper around two integers, but those two integers are the lever for the entire speed/quality trade-off in a Wan 2.2 graph. If your workflow "feels slow," this is the first place to look - dropping accelerated_steps from 10 to 8 changes every accelerated stage in the graph, which is the kind of single-point control the rest of the pack's helper selectors exist to give you.