ControlNet Custom Weights [Flux] ππ π π
Per-block Flux ControlNet control
- cn_extras
- CN_WEIGHTS
- TK_SHORTCUT
This is the per-block weights node built specifically for Flux ControlNets. Flux isn't a UNet - it's a DiT, so its ControlNets hook into transformer blocks rather than the SD-style encoder blocks. The block count and layout are different, which is exactly why the SD1.5 custom-weights node doesn't fit here and this Flux-specific one exists. If you're running a Flux ControlNet and want to reweight it block by block, this is the node.
It exposes 19 inputs (input_0 through input_18), one strength per block, all defaulting to 1.0 - uniform full control out of the box. You touch them only when you want to bias which blocks the ControlNet leans on.
Do you actually need this?
Honest answer: most Flux ControlNet workflows never touch per-block weights. The ControlNet KB's guidance for the modern union models is to start at the published conditioning scale - Alibaba PAI's FLUX.2 union recommends roughly 0.65β0.80 overall, and starting at the old 1.0 default overcooks it - and that overall strength dial (on the Apply node) is the knob that matters 95% of the time. This node is for the remaining 5%: when a global strength adjustment isn't enough and you want to shape where in the network the control applies. It's a precision tool, not a default.
Inputs and outputs
- input_0 β¦ input_18 (19 required) - the per-block strengths for a Flux ControlNet, default 1.0 each.
- uncond_multiplier (optional) - 0.0 replicates "ControlNet is more important" (drops the ControlNet from the unconditional pass); between 0 and 1 scales it continuously. Default 1.0.
- cn_extras (optional) - chain in extra weight modifiers.
Outputs: CN_WEIGHTS - into a Timestep Keyframe's cn_weights, or weights_override on Apply Advanced ControlNet - and TK_SHORTCUT, a Timestep Keyframe you can wire straight into a timestep_kf input.
Installing the pack
ComfyUI Manager β Install Custom Nodes β search "ComfyUI-Advanced-ControlNet" β restart. Or cd ComfyUI/custom_nodes && git clone https://github.com/Kosinkadink/ComfyUI-Advanced-ControlNet then restart. Pure ComfyUI, Kosinkadink's pack. The pack has no preprocessors (comfyui_controlnet_aux) and no ControlNet checkpoints - download a Flux ControlNet (a union like Shakker's or Alibaba PAI's is the usual pick) into ComfyUI/models/controlnet.
Common gotchas
Architecture match first. These 19 block weights are for Flux ControlNets. Point this node at an SD 1.5 ControlNet and the weights won't map - you'll get a compatibility error naming the weight types the loaded ControlNet actually accepts. Use the [SD1.5] node there.
Second, don't reach for per-block Flux weighting before you've tuned the basics. On a Flux union the biggest lever is the overall conditioning scale and the start/end step window on the Apply node; the KB is clear that the 2025β26 unions want lower strengths than the SD-era 1.0 default. Get those right, and you'll usually find you never needed to touch these 19 sliders at all. When you do, change a few blocks at a time and watch what moves - nineteen simultaneous edits are impossible to reason about.
Inputs (21)
| Name | Type | Default | Description |
|---|---|---|---|
| input_0 | FLOAT | 1.0000β10 | β |
| input_1 | FLOAT | 1.0000β10 | β |
| input_2 | FLOAT | 1.0000β10 | β |
| input_3 | FLOAT | 1.0000β10 | β |
| input_4 | FLOAT | 1.0000β10 | β |
| input_5 | FLOAT | 1.0000β10 | β |
| input_6 | FLOAT | 1.0000β10 | β |
| input_7 | FLOAT | 1.0000β10 | β |
| input_8 | FLOAT | 1.0000β10 | β |
| input_9 | FLOAT | 1.0000β10 | β |
| input_10 | FLOAT | 1.0000β10 | β |
| input_11 | FLOAT | 1.0000β10 | β |
| input_12 | FLOAT | 1.0000β10 | β |
| input_13 | FLOAT | 1.0000β10 | β |
| input_14 | FLOAT | 1.0000β10 | β |
| input_15 | FLOAT | 1.0000β10 | β |
| input_16 | FLOAT | 1.0000β10 | β |
| input_17 | FLOAT | 1.0000β10 | β |
| input_18 | FLOAT | 1.0000β10 | 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 |