IPAdapter Layer Weights Slider

IPAdapter Layer Weights Slider is the one genuinely original node in this pack - it's the feature chflame163 added on top of the copied codebase, and it exists to make one of the most confusing parameters in IPAdapter Mad Scientist V2 actually usable. The Mad Scientist takes a layer_weights string - a list like 0:1.0, 1:0.5, 2:0.0 that scales how strongly the adapter's influence is applied at each attention layer of the model. In theory that's powerful. In practice, typing those strings by hand and guessing is miserable. This node gives you twelve sliders instead.

How it works

layer0 through layer11 are the sliders, each defaulting to 0 and ranging −1 to 1 in 0.01 steps - one per attention layer (the adapter's influence spans 12 layers in the model's block structure). Negative values are allowed, which lets you actively suppress the image influence at a given layer rather than just zero it. multiplier (default 1, up to 999) scales the whole set: the sliders are capped at ±1, so if your workflow calls for a layer weight of 2.0 you set the slider to 1 and the multiplier to 2. The output is a single layer_weights STRING, formatted exactly the way the Mad Scientist expects (0:x, 1:y, ...), so you can wire it straight into that node's layer_weights input.

When it's worth your time

The honest take: this is a dev-tier tool for people who understand what per-layer control buys them. The classic use is the style/detail split - early layers tend to carry composition and layout influence, later layers carry texture and fine detail, and layer weights let you emphasize one over the other. But if you're not already getting the results you want from weight, start/end, and weight_type, this won't rescue you; it's a refinement knob on top of a working setup. The slider form is the whole value: it turns an opaque string format into something you can grab and drag, which is how you actually discover what the layers do.

Installing it

ComfyUI Manager: search "ComfyUI_IPAdapter_plus_V2". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/chflame163/ComfyUI_IPAdapter_plus_V2

Restart ComfyUI. No model files needed - pure parameter generation.

Gotchas

The sliders and the multiplier multiply, and the result is passed verbatim into the layer weights string, so sanity-check the Mad Scientist output when you start pushing values. Also note this outputs a string, not a numeric list - if you're trying to wire it into a Batch apply node instead of the Mad Scientist, it won't fit. And while the pack's README highlights this node, the pack itself remains a renamed snapshot of cubiq's V2 codebase, so everything else in it tracks that state.