Get sigmas as float

The adapter node nobody thinks about until they need it

Most of sigmas_tools_and_the_golden_scheduler deals in SIGMAS tensors - the lists of noise levels that flow into samplers. Get sigmas as float is the exception: it takes a SIGMAS tensor and collapses it into a single FLOAT. That one number is the noise level of the first step, in the units a latent actually uses, which is exactly what the README says it's for: "get first - last step to be able to inject noise inside a latent with noise injection nodes."

Here's the context. Noise injection nodes (the ones that add noise into a latent before or between sampler passes) often want an actual noise magnitude as a plain number, not a schedule. But the raw sigma value from your scheduler is in model units - and it needs scaling by the model's latent_format.scale_factor (that's the 0.18215-style constant baked into SD-family models, and a different one for flow-matching models). This node does that math for you:

sigfloat = (sigmas[0] - sigmas[-1]) / scale_factor

Since the last sigma in a normal schedule is 0, you're effectively reading the first step's sigma, rescaled into noise-injection units.

What you set

Two inputs:

• model - needed so the node can read that model's scale_factor
• sigmas - any SIGMAS tensor

One FLOAT output, which you feed into whatever noise-injection node is asking for a noise level.

The honest take

This is a narrow, purpose-built adapter - you won't use it in every workflow, but when you're doing manual noise injection and find yourself hardcoding a noise value that never quite matches what the sampler expects, it's the missing link. It also makes for a nice diagnostic: wire it to a display and you can read off how "loud" the start of any schedule actually is, in comparable units across models.

One caveat, straight from the source: if the model has no scale_factor (or it's 0), the node prints a warning and returns 0.0 rather than crashing - so a 0 output usually means "this model doesn't expose a scale factor," not "your schedule is silent."

Install & gotchas

ComfyUI Manager → search sigmas_tools_and_the_golden_scheduler, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Extraltodeus/sigmas_tools_and_the_golden_scheduler

Restart, no model downloads. Pack-level gotcha: requirements.txt lists only asteval while the code imports matplotlib and scipy, so pip install matplotlib scipy asteval if the pack won't load on a minimal install.