✏️ Literal Float

✏️ Literal Float holds one decimal number and outputs it. value in, value FLOAT out - a straight pass-through. The range is generous (about -10 billion to +10 billion) and the step is 0.001, which covers CFG scales, denoise strengths, LoRA weights and upscale factors without arguing.

You use it the same way you use any value node: right-click a float widget somewhere - cfg on a KSampler, denoise on an image-to-image pass, the strength on a LoRA loader - pick Convert widget to input, and now that widget is a socket. Drop this node on it and the number has one home.

Where it earns its place

The reason that's worth a node: denoise and CFG are the two settings you tune most and forget to keep consistent. Two sampler passes in one graph, each with its own denoise box, is how you end up wondering why your hires pass re-rolled the composition. One Literal Float feeding both, and changing it is one edit. Same story for CFG across a base pass and a refinement pass.

There's a second, quieter use: exposing a number you only ever change in a certain window. A LoRA strength you nudge constantly is easier to find as a labelled node on the canvas than as a widget buried in a loader's third row.

What it won't do

It doesn't do math, it doesn't randomise, and it doesn't clamp anything downstream - those are different nodes (the same pack has Random Float in Range if you want variation per run). And converting a widget to an input means the widget is gone: you can't type into it directly any more, which occasionally surprises people who wire something up and then want to override one node by hand. Disconnect the wire to get the box back.

Worth saying plainly: core ComfyUI now has its own typed primitives, so this isn't a capability you can only get from this pack. The pack version is just a conventionally named float source sitting in the same category as everything else you'll add from MNeMiC.

Install

ComfyUI Manager → ComfyUI-mnemic-nodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/MNeMoNiCuZ/ComfyUI-mnemic-nodes

Restart afterwards. No downloads for this node, though the pack's shared requirements.txt installs a full Python ML dependency set (transformers, opencv-python, tiktoken, imageio, piexif, groq…) because it's one bundle for ~40 nodes. That's the standing cost of utility packs: you wanted a float, you got a toolchain.