VarBoard Float

The floats are where most of your actual tuning happens. CFG scale, denoise, LoRA strength, IPAdapter weight, upscale factor - stock ComfyUI scatters them across a dozen nodes, and you end up memorizing where each one lives. VB_Float pulls those decimal dials onto the VarBoard panel so they sit side by side.

What it does

Like every node in this pack, the backend is a pass-through: execute() returns (value, label) untouched. The value lives in the node's widget, and the frontend renders a slider row on the floating panel anchored by VB_Panel. Move the slider on the board and it writes back into the widget, so the next run feeds the new number to whatever you connected.

The inputs are the two you'd expect:

• value - the float. Default 0.00, range -10.00 to 10.00, step 0.01.
• label - the name shown on the board. Default "Float". Call it "CFG", "Denoise", "LoRA strength" - that's what you'll read while working.

Outputs: FLOAT (wire into your target's float input) and LABEL (the label string).

Where the default range bites

Here's the trap worth knowing before you build a workflow around this node: the default range caps at 10.00. Denoise and most LoRA strengths live happily inside that, but CFG scale routinely wanders past 10, and if you like strong guidance you will hit the ceiling. It's not a hard blocker - the panel supports per-node range overrides in its settings, so you can widen this node's span from the board without touching the graph - but if you hit a slider that stops at 10 and won't go further, now you know why.

Install

No dependencies, no models, nothing beyond a working ComfyUI. ComfyUI Manager (search "VarBoard"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/IA-gyz/comfyui-VarBoard

Restart ComfyUI, hard-refresh the browser, drop a VB_Panel anchor, and your VB_Float nodes appear on it automatically.

When you'd actually use it

This is the node that makes sweep-testing pleasant. Keep a fixed seed, set up a couple of VB_Float nodes for CFG and denoise, and you can flip between runs without ever zooming the canvas. That fixed-seed discipline is the whole game in troubleshooting - change one variable at a time or you can't tell what moved the image. A panel that puts the two variables you're testing right next to each other is the difference between actually doing that and giving up halfway.