🅐 Float

There is a category of ComfyUI node so boring it never gets a blog post, yet you'd notice the moment it's missing. 🅐 Float is one of those. It's a single number widget wrapped in a node: you type a float, you get a float out. That's the whole spec. But "the whole spec" is exactly why people reach for it, because the pack's README frames these primitives as "nodes not included in ComfyUI core yet" - and there's real truth to that for anyone who's ever wrestled with the built-in primitive.

Why you'd bother

The stock ComfyUI Primitive node is a loose, any-typed value. It works, but it's also the source of a classic beginner faceplant: it silently converts types, and a "number" that's actually being treated as something else can flow into a node and produce confusing output. A typed FLOAT port like this one can only ever be a float, so the type system does the arguing for you - if it connects, it's a float.

Concretely, this is your denoise knob, your CFG tweak, your strength slider for a LoRA or IP-Adapter weight. The input is a standard FLOAT widget with a fine 0.001 step and a range wide enough to cover anything sane, and the output is a clean FLOAT you can wire into any node that wants one. Because the widget sits right on the node, you can nudge it between runs without digging into a nested primitive panel - which is the small quality-of-life win utility packs like this are actually selling.

Inputs and outputs

• float - FLOAT, default 0.0, step 0.001.
• Output: FLOAT.

Nothing else. It doesn't clamp, convert, or compute; it passes the value through and labels it correctly. If you need math on top of it, that's what the pack's 🅐 Expression Eval is for - wire this float into the a or b port and compute downstream.

Install

Part of the hay86/ComfyUI_AceNodes pack:

cd ComfyUI/custom_nodes
git clone https://github.com/hay86/ComfyUI_AceNodes
cd ComfyUI_AceNodes
pip install -r requirements.txt

Restart ComfyUI and you'll find 🅐 Float under Ace Nodes. ComfyUI Manager users can search the pack title ComfyUI_AceNodes and install from there.

The honest caveats

This node's only real downside is the pack's install footprint. pip install -r requirements.txt drags in a long list of heavy packages - rembg, transformers, insightface, retina-face, boto3 - none of which this primitive touches. If you only want the primitives, you're paying for the whole pack's dependency list anyway. That's a pack-level tradeoff, not a node-level one, but it's worth knowing before you install.

Also note the distinction the pack draws: this is a typed float, not a seed or an integer. Keep it for continuous values - the moment you need a whole number, grab 🅐 Integer instead and let the type system keep you honest.