sanmi_Float

A pass-through node: you type a value, it hands you that same number as a FLOAT output. Nothing is computed, nothing is transformed. This is the ComfyUI equivalent of a blank index card - and before you roll your eyes, blank index cards exist because sometimes you need them.

What it's for

ComfyUI already ships primitive nodes, but this pack's author added their own set so that workflows built around the pack's other "Sanmi Simple Nodes" would work standalone without core nodes. In practice a Float passthrough gets used for three things: converting a typed number into a clean FLOAT-typed output to wire into nodes that only accept that type; exposing a tunable constant (CFG, denoise, strength) that you can adjust without digging into the node that consumes it; and breaking an unwanted type connection when a node's strictness is getting in the way. It's also a handy placeholder - drop it in a spot in a template workflow and it's obvious where a value is meant to go.

Inputs and outputs

One input: value (FLOAT, default 0, step 0.1). One output: float (FLOAT). That's the whole surface area - there is genuinely nothing else to configure.

Why you might not need it

Core ComfyUI gives you a Float input node out of the box, and dozens of utility packs ship the same thing with nicer defaults. The only reason to use this one is if you're standardizing on this pack's node set (or you found it in a downloaded workflow that references it, in which case you need it to keep the graph loadable). And there's the standing caveat for everything here: the pack was rewritten in April 2025 to contain only the GPT4V captioner, so this node no longer ships in a fresh clone - workflows referencing it will throw a "node not found" until you restore the old Load_Image_Batch.py from the repo's history.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/438443467/ComfyUI-GPT4V-Image-Captioner

Restart ComfyUI. No models, no dependencies. If the node is missing after install, that's the April 2025 trim - not a broken install.