ControlNet Soft Weights ππ π π
Replicate 'My prompt is more important' from Auto1111
- cn_extras
- CN_WEIGHTS
- TK_SHORTCUT
If you ever used Auto1111's sd-webui-controlnet extension, you'll remember the "My prompt is more important" control mode - a way to keep ControlNet's structure guidance without letting it steamroll your prompt. SoftControlNetWeights is that exact behavior, ported to ComfyUI as a standalone weights node: instead of one flat strength applied to every ControlNet layer equally, it applies a per-layer curve that tapers off, giving the prompt more say the deeper into the network you go.
How it works
A standard SD1.5/SDXL-style ControlNet has thirteen internal blocks, and this node gives you one weight slider per block - weight_00 through weight_12. Left at the defaults, they form a soft ramp: roughly 0.10 β 0.12 β 0.15 β 0.18 β 0.21 β 0.26 β 0.32 β 0.38 β 0.46 β 0.56 β 0.68 β 0.83 β 1.0. That's the "soft" curve - light influence early, building to full strength only at the last block - which is what gives the prompt room to breathe rather than being overridden outright. Most people never touch the thirteen sliders individually; they exist for fine control, but the shipped default is the point of the node.
flip_weights reverses that curve - front-loaded instead of back-loaded - if you want the opposite emphasis. uncond_multiplier is the other half of the Auto1111 feature pair: it's what "ControlNet is more important" maps to. Set it to 0.0 and you get Auto1111's behavior exactly; anything between 0 and 1 gives you a granular version the original extension didn't offer.
The inputs and outputs that matter
weight_00βweight_12- the thirteen per-block values (0β10 each, defaults form the soft decay curve above). Leave these alone unless you're deliberately hand-tuning the curve.flip_weights(default false) - inverts the curve direction.uncond_multiplier(optional, default 1.0) - 0.0 reproduces "ControlNet is more important" exactly; values between granularly dial it in.cn_extras(optional) - a side channel for architecture-specific extras, documented in the README for Anima LLLite's inpainting mask.
Outputs: CN_WEIGHTS feeds a Timestep Keyframe's cn_weights input (or Apply Advanced ControlNet's weights_override, which the README is explicit applies to every timestep regardless of any other schedule you've built - only use it if that's genuinely what you want). TK_SHORTCUT packages the same weights as a ready-made Timestep Keyframe, so you can skip building one separately if you don't need any of its other features.
How to install it
Through ComfyUI Manager: search "ComfyUI-Advanced-ControlNet", install, restart. Manually:
cd ComfyUI/custom_nodes
git clone https://github.com/Kosinkadink/ComfyUI-Advanced-ControlNet
then restart ComfyUI. No model downloads and nothing heavy to install - it's a pure weighting node sitting on top of whatever ControlNet you've already loaded.
Common issues & troubleshooting
"weight types are compatible" error. This node is shaped for the standard 13-block SD1.5/SDXL ControlNet architecture. If you loaded a T2I-Adapter, this won't validate - swap to SoftT2IAdapterWeights instead, which matches that architecture's shallower structure.
No visible difference from a flat strength. The soft curve is a subtle effect, not a dramatic one - it's meant to rebalance prompt-vs-structure at the margins, not gut ControlNet's influence. If you want a stronger version of the same idea, push uncond_multiplier toward 0 rather than hand-editing the individual weight sliders.
Update issues via ComfyUI Manager. This pack has shown up in community reports where Manager claims "you're on the latest version" when a real update is available. If behavior doesn't match current docs, cd into the pack's folder inside custom_nodes and run git pull manually before assuming something's broken.
Inputs (16)
| Name | Type | Default | Description |
|---|---|---|---|
| weight_00 | FLOAT | 0.0990β10 | β |
| weight_01 | FLOAT | 0.1210β10 | β |
| weight_02 | FLOAT | 0.1460β10 | β |
| weight_03 | FLOAT | 0.1770β10 | β |
| weight_04 | FLOAT | 0.2150β10 | β |
| weight_05 | FLOAT | 0.2600β10 | β |
| weight_06 | FLOAT | 0.3150β10 | β |
| weight_07 | FLOAT | 0.3820β10 | β |
| weight_08 | FLOAT | 0.4630β10 | β |
| weight_09 | FLOAT | 0.5620β10 | β |
| weight_10 | FLOAT | 0.6810β10 | β |
| weight_11 | FLOAT | 0.8250β10 | β |
| weight_12 | FLOAT | 1.0000β10 | β |
| flip_weights | BOOLEAN | false | β |
| uncond_multiplieropt | FLOAT | 1.000β1 | β |
| cn_extrasopt | CN_WEIGHTS_EXTRAS | β |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| CN_WEIGHTS | CONTROL_NET_WEIGHTS | β |
| TK_SHORTCUT | TIMESTEP_KEYFRAME | β |