List Custom Float

Float versions of "give me a list of numbers" are the unglamorous workhorse of workflow tuning. CFG scales to sweep, denoise strengths per pass, weights for a blend, a ramp of values to try - all of those want a list of floats, and stock ComfyUI makes you build them one widget at a time. This node takes a single multiline text box and turns it into a float list, with the same range-and-stride syntax as the pack's int version, plus decimals.

How it works

Type into custom_text, get float_list out the other end, with a count INT alongside. The syntax is the useful part:

• Plain values with decimals: 0.2, 0.4, 0.6
• Bracket ranges with stride: [0.5, 2.0, 0.25] → 0.5, 0.75, 1.0, ... 2.0
• Dash ranges: 1.0-2.0:0.1 → 1.0, 1.1, 1.2, ... 2.0
• Mixed and messy input is fine - the parser walks through, expanding ranges as it goes.

It normalizes full-width Chinese punctuation (，；：（）【】) to ASCII, which sounds niche until you're copying a parameter list from a Chinese-language workflow and it just works. Empty input falls back to [0.0].

Outputs: float_list (a list-type output, so list-aware downstream nodes see a proper list) and count. Feed the list into anything that takes a sequence of floats - or pair it with a batch/loop pattern to run one sampler at several CFG values and compare.

Where you'd use it

Two patterns dominate. First, sweeps: build [7, 7.5, 8, 8.5, 9] once, drive a KSampler's CFG through it, and compare results without touching widgets between runs. Second, ramps: denoise strengths like 0.4, 0.3, 0.2, 0.1 for a progressive refine chain, where the order is the whole point. The count output is handy when a loop needs to know how many values it's iterating over.

Installing it

Part of 1hewNodes. ComfyUI Manager → search "ComfyUI 1hewNodes" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/1hew/ComfyUI-1hewNodes

Restart ComfyUI. No special dependencies, no model downloads.

Common issues

Same traps as the int version, with one float-specific addition: bracket and dash ranges are inclusive on both ends, so [0.5, 2.0, 0.25] ends exactly at 2.0 - if you expected an open-ended ramp you'll get an extra value. And because it's text parsing, a stray decimal or a Chinese comma that slipped past normalization can quietly add or drop an item; glance at the count output when the list looks short. Remember float_list is a list, not a batch - if a downstream expects a tensor batch, run it through a list→batch conversion first.