Float 10.00

You installed the Rebalance Pack for the Krea 2 conditioning node and now there are forty nodes in your menu. Float 10.00 is one of the small ones - a single number you type, with a wire coming out. It sounds pointless until you're ten iterations into a CFG sweep and can't tell whether your two "identical" runs actually were.

What it's for

It's a stand-alone float: one value in, the same value out. The reason a node like this exists at all is Convert widget to input. Right-click the CFG or denoise widget on a KSampler, convert it to a socket, and now that value has exactly one source instead of being retyped on every sampler in the graph. Core ships PrimitiveFloat for the same job; this one is the pack's take on it, with two differences that matter: the range is locked to ±10 and the value is cleaned up before it leaves the node.

That range is a decent fit for the dials people actually sweep. CFG, denoise, strength, most LoRA weights - they all live inside ±10, and a slider that physically stops at 10.00 is a guardrail against the stray 100 that turns a sampler into a lava lamp. If your number lives outside that window, you want Float Exact from the same pack instead.

How it works

The input is a FLOAT with min: -10.00, max: 10.00, step 0.01. On the way out the node does one thing:

return (round(value, 2),)

That's the entire node. The rounding is the interesting half. ComfyUI's frontend works widget steps in binary float math, so a value the UI displays as 0.30 can arrive at the node as 0.30499999999999994. It gets written into your workflow JSON, it gets passed downstream, and it explains a class of bug reports where "the same settings" produce measurably different output. Snapping to the 0.01 grid kills it. The author's own comment in the source is exactly that, with that exact example.

So: same value in, same value out, minus the float dust.

Inputs and outputs

• value - the FLOAT widget. Default 0.0, range -10.00 to 10.00, step 0.01.
• FLOAT - the output. One socket, connect it to any float input you've converted to a wire.

Installing it

It's part of the Rebalance Pack, which means one install for this node and the other thirty-nine.

cd ComfyUI/custom_nodes
git clone https://github.com/nova452/Rebalance-Pack.git
# restart ComfyUI

Or search Rebalance Pack in ComfyUI Manager (publisher nova452) and install. The repo was renamed from ComfyUI-ConditioningKrea2Rebalance, so older links use that name - GitHub redirects to the same place. No requirements.txt, no install script, no models to download; everything it imports (torch, numpy, Pillow) already ships with ComfyUI.

Where it bites

It rounds to two decimals, and it does that silently. Wire Float 10.00 into something that genuinely wants three digits - a LoRA weight of 0.125, a scale factor that moves in steps of 0.005 - and you'll get 0.13 out. The node is honest about it in the source and dishonest about it in the UI; nothing warns you. That's the trade for the cleanup, and it's the one case where Float Exact is the correct pick.

The clamp is real, so don't use it for anything unconventional. Negative denoise, big model-merge multipliers, sigmas: not this node. It'll pin you at ±10 and you'll spend twenty minutes wondering why your value won't take.

Duplicate sources of truth do more damage here than the float dust does. The float artifacts this cleans up are cosmetic most of the time. What actually breaks workflows is three nodes that each look like they set the same value. Use it once, wire it everywhere, and note that the pack's other primitive nodes (Float 1.00 for 0–1 ranges, Any for wildcards) exist for the same reason.