Nodes/ComfUI-EGAdapterMadAssistant/IPAdapter Mad Assistant V4
ComfyUI Node

IPAdapter Mad Assistant V4

-4 to 4, for the rare layer that needs everything

By 11dogzi·Created 2 years ago·Updated 2 years ago· 40
IPAdapter Mad Assistant V4
    • layer_weights
    0layer Weight0.0
    1layer Weight0.0
    2layer Weight0.0
    3layer Weight0.0
    4layer Weight0.0
    5layer Weight0.0
    6layer Weight0.0
    7layer Weight0.0
    8layer Weight0.0
    9layer Weight0.0
    10layer Weight0.0
    11layer Weight0.0
    RandomNo
    seed0

    IPAdapter Mad Assistant V4 is the same slider-to-string helper as the rest of the family, with the twelve layer Weight sliders stretched to -4.0 to 4.0. It's another rung on the same ladder: each V-number just widens the range, and V4 is where the ladder stops feeling like extra headroom and starts feeling like overkill.

    To be direct: you will almost never need a weight this large. The Mad Scientist ecosystem's honest advice is that a 0.7–0.8 weight already lets the reference override the prompt, and the per-layer versions exist so you can push a specific layer further without nuking the whole image. At 4.0 on one layer, you're asking that layer to reproduce the reference's signal at maximum force - a legit trick for a late style layer, paired with early structure layers parked near zero, but a very narrow job. The realistic V4 workflow has one slider at the extreme and the other eleven barely off zero, and you arrived here by testing V2 and V3 first.

    Nothing else changes. Twelve sliders in, comma-joined layer:weight string out through the layer_weights STRING output, wired into the IPAdapter Mad Scientist node's layer_weights input after converting that widget to an input. Random mode fills zero layers from the full -4.0 to 4.0 range - at this spread it's a dice roll with real teeth, so keep non-zero layers pinned and probably don't use it here at all. The seed input is declared in the schema but never referenced by the code, so random mode isn't reproducible no matter what you set.

    Which one do you want?

    V4 is the answer to "I tried V3 and the extreme still wasn't extreme enough." If you're at this point, you already know which layer you're fighting, and this node is just permission to fight it harder. For everyone else, the base node (0..1) or V1 (-1..1) is the daily driver.

    Install

    Same single-file pack, same steps:

    cd ComfyUI/custom_nodes
    git clone https://github.com/11dogzi/ComfUI-EGAdapterMadAssistant
    

    Restart ComfyUI, or search ComfUI-EGAdapterMadAssistant in ComfyUI Manager. No pip, no dependencies, no model files - this pack ships zero requirements. You still need ComfyUI_IPAdapter_plus and its models for the Mad Scientist node to exist.

    If the output seems to do nothing, the usual suspect applies: you forgot to convert Mad Scientist's layer_weights widget to an input before wiring this node into it.

    Categoryipadapter

    Inputs (14)

    NameTypeDefaultDescription
    0layer WeightFLOAT0.0-4–4
    1layer WeightFLOAT0.0-4–4
    2layer WeightFLOAT0.0-4–4
    3layer WeightFLOAT0.0-4–4
    4layer WeightFLOAT0.0-4–4
    5layer WeightFLOAT0.0-4–4
    6layer WeightFLOAT0.0-4–4
    7layer WeightFLOAT0.0-4–4
    8layer WeightFLOAT0.0-4–4
    9layer WeightFLOAT0.0-4–4
    10layer WeightFLOAT0.0-4–4
    11layer WeightFLOAT0.0-4–4
    RandomCOMBONo2 options: Yes, No
    seedINT0-1125899906842624–1125899906842624

    Outputs (1)

    NameTypeDescription
    layer_weightsSTRING