LogicUtil_Random Shuffle Float

Want a menu of strengths handed out in a different order every run, but still repeatably? LogicUtil_Random Shuffle Float splits a list, shuffles it with a seeded RNG, and hands back the reordered result. It's the shuffle member of the family, and the seed is what makes it actually usable: change the seed, get a new order; keep the seed, get the same order every time.

How it works

Three inputs: input_string (your list as one line, default 1.0$2.0$3.0), separator (default $), and seed (an INT from 0 to 9999999999, default 0). The node splits on the separator, runs the pieces through random.Random(seed).shuffle(), and outputs the shuffled result.

Now the two things that will trip you up. First, despite the "Float" in the name, nothing is cast to float. The pieces are shuffled as strings and come out as strings. 1.0$2.0$3.0 shuffled might come back as ["2.0", "1.0", "3.0"] - text, not numbers. The Int and String variants behave identically; the name is aspirational. Second, the output is a list, not a single string - the schema declares STRING, but what actually flows out is the shuffled list, which is how the LogicUtil family threads lists between nodes. Feed it to a list-aware consumer, not to a field that wants one value.

Where it earns its keep: randomized-but-reproducible ordering. Shuffle a list of prompt styles so each queue run gets a fresh order, but keep the seed fixed when you're debugging so you can reproduce the exact run - the community's "change one variable at a time with a fixed seed" rule applies to ordering too. It also pairs naturally with the Manual Choice nodes: shuffle the list, then index into the shuffled order for "random without repetition" style behavior.

It's part of the LogicUtil family - a port of aria1th's ComfyUI-LogicUtils riding inside the JDCN pack from Daxton Caylor and Jerry Davos, whose README documents the file and latent batch nodes and never mentions this family. When the graph shows LogicUtil_Random Shuffle Float, that's the original class name with the pack prefix.

Install

Install is the standard JDCN routine: ComfyUI Manager → "Install Custom Node" → search JDCN → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/daxcay/ComfyUI-JDCN
cd ComfyUI-JDCN && pip install -r requirements.txt

Only piexif on the requirements list, no models to fetch. And if your "shuffled floats" come out as strings and a downstream node complains, that's not a bug - cast them with a conversion node, or reach for the Int/Float choice nodes instead.