ComfyUI Node

SettingsSwitch

Three saved presets for the ipadapter node's four dials

By taabata·Created 3 years ago·Updated 2 years ago· 259
SettingsSwitch
    • FLOAT
    • FLOAT
    • FLOAT
    • FLOAT
    sf_10.60
    sf_20.60
    sf_30.60
    strength_10.60
    strength_20.60
    strength_30.60
    IPAScale_10.60
    IPAScale_20.60
    IPAScale_30.60
    CNScale_10.60
    CNScale_20.60
    CNScale_30.60
    setting

    SettingsSwitch is a utility that exists to make the pack's fanciest node testable. The LCMLora_ipadapter generate node has four conditioning dials - style fidelity, strength, IP-Adapter scale, and ControlNet scale - and tuning all four by hand, then re-tuning them for a second style, is the kind of fiddling you want to do fast. This node banks three full presets of those four values and flips between them with a single setting dropdown (v1, v2, v3). Pick the bank, and all four of its outputs follow.

    It does no generation, takes no images. It's a four-way float router with a three-way selector, and that's exactly what it should be.

    How it works

    Twelve float inputs in three groups of four. Each group (_1, _2, _3) holds one complete setting for the four dials:

    • sf - style fidelity.
    • strength - generation strength.
    • IPAScale - IP-Adapter scale.
    • CNScale - ControlNet scale.

    The setting enum picks a group, and the node emits that group's four values in order on four FLOAT outputs. That's the entire mechanism - a lookup table with a switch on top.

    The inputs and outputs that matter

    • setting - v1 / v2 / v3. The only input you'll touch while testing.
    • The twelve floats - set up once per workflow: sf_1..3, strength_1..3, IPAScale_1..3, CNScale_1..3, all defaulting to 0.6.

    Outputs (in order): FLOAT ×4 - sf, strength, IPAScale, CNScale. Wire them in that order into LCMLora_ipadapter's style_fidelity, strength, ipadapter_scale, and controlnet_conditioning_scale inputs. All four ranges are 0–1 with 0.01 steps, which lines up with the generate node's style-fidelity and strength fields (its ipadapter/controlnet scales technically accept up to 10, but the switch caps at 1.0 - fine for real use).

    How to install it

    Part of the pack:

    cd ComfyUI/custom_nodes
    git clone https://github.com/taabata/LCM_Inpaint-Outpaint_Comfy
    cd LCM_Inpaint-Outpaint_Comfy
    pip install -r requirements.txt
    

    Or ComfyUI Manager → "LCM_Inpaint-Outpaint_Comfy" → restart. No models, no dependencies beyond what the pack already pulls in.

    Common issues

    There's not a lot to break, but two things are easy to get wrong. First, the output order is fixed and unlabeled beyond position - sf, strength, IPAScale, CNScale. If you wire the four floats out of order, the pack's ipadapter node silently misbehaves, and it'll look like a tuning problem when it's actually a crossed wire. Second, all values cap at 1.0 here, and the generate node's ipadapter_scale/controlnet_conditioning_scale accept up to 10 - so if you ever need aggressive scales above 1.0, this switch physically can't hold them and you'll have to type them in manually instead.

    It's a small node, but it's genuinely the right tool for its one job: A/B/C testing four interdependent dials without rebuilding the graph. Set each bank once, and flipping between looks becomes one dropdown instead of four sliders.

    Categoryimage

    Inputs (13)

    NameTypeDefaultDescription
    sf_1FLOAT0.600–1
    sf_2FLOAT0.600–1
    sf_3FLOAT0.600–1
    strength_1FLOAT0.600–1
    strength_2FLOAT0.600–1
    strength_3FLOAT0.600–1
    IPAScale_1FLOAT0.600–1
    IPAScale_2FLOAT0.600–1
    IPAScale_3FLOAT0.600–1
    CNScale_1FLOAT0.600–1
    CNScale_2FLOAT0.600–1
    CNScale_3FLOAT0.600–1
    settingCOMBO3 options: v1, v2, v3

    Outputs (4)

    NameTypeDescription
    FLOATFLOAT
    FLOATFLOAT
    FLOATFLOAT
    FLOATFLOAT