Float Switch

The Float Switch is the pack's standard two-position toggle applied to numbers. An Input widget set to 1 or 2 decides whether input1 or input2 flows out of the float output. No conversion, no clamping, no rounding - it picks a wire and hands it through. If you've ever wanted to flip between a fast draft denoise (0.5) and a final refine denoise (0.25) without rebuilding the graph, this is the node.

How to actually use it

The classic setup: two preset floats feeding the switch, and the switch's output driving a KSampler's denoise port (or a ControlNet weight, or a guidance strength - anything float-shaped). To switch, you move the Input widget from 1 to 2. That's the whole interaction, and it's the fastest possible A/B test for "does a stronger denoise fix this?"

The inputs are forceInput, so the values can't be typed into the node - they must come from upstream: the pack's own Float primitive, a math node, or another switch. That's a feature in disguise: it means the switch is routing wired values, so the same two presets can be shared across multiple switches or nodes.

One genuinely useful pattern: leave input2 empty. Set Input to 2 with nothing connected, and the switch passes None. Depending on what's downstream, None often means "use the node's own default" or gets treated as "skip" - for a denoise feed it's usually safer to wire an explicit value, but for optional strength slots it's a clean way to express "off."

And remember the pack-wide convention from the README: if the switch node is bypassed (not muted), it sends whatever's on input1. Bypass = "default to slot 1". So the value you want when someone bypasses the whole switch belongs on input1.

What's on the node

• Input - the 1-or-2 selector widget.
• input1 / input2 - optional FLOAT inputs, wire-only.
• float output - the chosen FLOAT.

If both inputs are empty you get None back, which the node will cheerfully output - float switches don't have the multi-switch's "Missing Input" guard. So wire them properly.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/r-vage/ComfyUI-RvTools_v2
# restart ComfyUI

Or via ComfyUI Manager, searching "ComfyUI-RvTools_v2". No models; the pack's deps are the standard torch/numpy/Pillow/opencv-python/pilgram.

The pack's standing caveat

The README declares RvTools v2 unmaintained and recommends ComfyUI_Eclipse as the successor. History worth knowing: the original RvTools repo was removed from GitHub in early 2025, breaking shared workflows and leaving a trail of "can anyone re-share RvTools?" posts on r/comfyui. v2 installs cleanly and is the version those legacy workflows expect. New builds: check Eclipse first.