T2IAdapter Custom Weights 🛂🅐🅒🅝

T2I-Adapters are the lightweight cousins of ControlNets - smaller, faster, lower VRAM, and the go-to when memory is the constraint (the ControlNet KB calls them "the lighter-weight alternative if VRAM is the constraint"). They also have far fewer injection points: just four. This node gives you one strength slider per those four blocks, so you can reweight a T2I-Adapter's guidance the same way the other custom-weights nodes do for full ControlNets.

Four inputs, not twelve or nineteen, because that's the architecture - a T2I-Adapter conditions the model at four feature scales rather than at every UNet block. This node matches that shape exactly, which is why you use it (and not the SD1.5 or Flux weights node) for T2I-Adapters.

When to use it

Honestly, T2I-Adapter workflows are usually "set the overall strength and go" - the whole appeal is being cheap and simple. You reach for per-block weights when a T2I-Adapter is either too weak overall or too dominant at one scale and a single global strength can't split the difference. The four blocks roughly span coarse-to-fine structure, so lowering the earlier ones loosens broad-composition grip while keeping finer guidance, and vice versa. It's a niche tool for a niche need, and that's fine.

Inputs and outputs

• input_0 … input_3 (4 required) - per-block strengths for the T2I-Adapter, default 1.0 each.
• uncond_multiplier (optional) - 0.0 for the "ControlNet is more important" behavior (drops it from the uncond pass), scaling continuously between 0 and 1. Default 1.0.
• cn_extras (optional) - chain in extra weight modifiers.

Outputs: CN_WEIGHTS (feeds a Timestep Keyframe's cn_weights, or weights_override on Apply Advanced ControlNet) and TK_SHORTCUT (a Timestep Keyframe to plug into a timestep_kf input directly).

Installing the pack

ComfyUI Manager: Install Custom Nodes → "ComfyUI-Advanced-ControlNet" → restart. Or cd ComfyUI/custom_nodes && git clone https://github.com/Kosinkadink/ComfyUI-Advanced-ControlNet and restart. Pure ComfyUI, no build step; it's Kosinkadink's pack. No preprocessors bundled (comfyui_controlnet_aux) and no model files - put your T2I-Adapter .safetensors in ComfyUI/models/controlnet.

Common gotchas

Use this node only with T2I-Adapters. Its four-block layout won't map onto a full ControlNet (which has many more injection points), and mixing them up throws a weight-type compatibility error. If you're on a full ControlNet, use the [SD1.5] or [Flux] custom-weights node.

Second, if you're new to the distinction: T2I-Adapters and ControlNets look interchangeable on the graph (both load into a ControlNet-style path and feed the same Apply node) but they're different model families with different block counts. When a weights node "doesn't fit," that mismatch is almost always the reason. And as with every weights node in the pack, these four values multiply against the overall strength on the Apply node.