Nodes/Sampling Planner/Acceleration Selector
ComfyUI Node

Acceleration Selector

The one-dropdown switch for Wan 2.2 acceleration modes

By boobkake22·Created 2 months ago·Updated 2 months ago· 0
Acceleration Selector
    • acceleration
    accelerationHigh + Low

    Acceleration Selector is the smallest kind of useful node: one dropdown, one output, and the entire point is that the dropdown lives somewhere visible instead of buried. It's a combo helper for Sampling Plan (Wan 2.2), mirroring the pack's Task Selector and Priority Selector. Pick a mode, feed the output to the plan's acceleration input, done.

    The four modes

    These map to which of Wan 2.2's two experts get a compatible acceleration LoRA:

    • High + Low (default) - both stages accelerated, the fast-but-fragile route
    • High only - accelerate the high-noise expert, keep low-noise at full quality
    • Low only - the community's compromise: full-quality high pass for motion and composition, speed LoRA only on the detail pass
    • None - no acceleration, the native full-quality profile

    The plan then does the real work: it projects your step budgets onto whichever stages are accelerated and reserves more steps for the unaccelerated one. This node just hands the choice over with a type-safe output that only connects where an acceleration value belongs.

    Why bother with a helper for one value?

    Two reasons, both the same as the other selectors in this pack. First, legibility: when you're hunting through a shared workflow for why it suddenly runs 4-step and plastic-skin-y, the answer is one visible dropdown, not a widget three panels deep. Second, fan-out: if the same mode switch also drives other branches - a bypass group, a CFG change, a different model loader - you wire them all from this one control instead of syncing multiple widgets by hand.

    One note on semantics: this node outputs the requested mode. The plan consumes it and produces the derived routing state (acceleration_state), which is what Acceleration Model Pair actually uses to decide which model branches execute. So think of the chain as: you pick the intent here, the plan interprets it, the model pair executes it. Keep the selector and the plan's own acceleration widget consistent - if you set one and the plan's own dropdown disagrees, you've got two sources of truth and the plan wins.

    Install

    It ships with the whole pack - ComfyUI Manager (search Sampling Planner) or:

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

    restart, and it's under sampling / Sampling Planner / Helpers. Nothing to download beyond the code.

    Honest verdict: if you never flip between accelerated and unaccelerated sampling, you don't need this node - set the plan's acceleration once and leave it. The moment you start A/B-testing "4-step speed vs 20-step quality" on the same workflow, a selector like this pays for itself in the first session.

    Categorysampling/Sampling Planner/Helpers

    Inputs (1)

    NameTypeDefaultDescription
    accelerationCOMBOHigh + LowConnect this output to the Sampling Plan acceleration input. The plan produces the derived routing state.

    Outputs (1)

    NameTypeDescription
    accelerationHigh + Low,High only,Low only,None