🧪 Sigma Plot (Experimental)

If you've been burned by the flow-model sampler mess - Karras wrecking a Wan or LTX render, or a workflow demanding a scheduler you've never heard of - BoyoExperimentalSigmaPlot is a breath of fresh air: a sigma schedule generator aimed squarely at flow/rectified-flow models (Wan, LTX, Krea2), with a picture of what it's about to do.

Here's the context that makes it make sense. Flow models like Wan and LTX are trained to predict a velocity along a near-straight line from noise to image, and they want their sigmas in the [0, 1] range - that's non-negotiable for them. The old DDPM-era schedulers assumed a different game, and aggressive reshaped schedules like Karras actively distort a straight flow trajectory instead of helping it. So a node that generates sane, visible [0, 1] schedules is genuinely useful - and one that draws a plot of them is a nice teaching tool on top.

Inputs, all four of them, are things you'll actually touch:

• steps - 4 to 12, default 8. The flow-model turbo range; step counts collapsed for these architectures, so don't crank it thinking more is better.
• sigma_min - default 0.03, and the tooltip says what matters: the BasicScheduler's simple schedule at 8 steps ends at 0.125, and lower values spend more time in fine detail. If you can't decide, that 0.03 default is a fine place to start.
• strategy - the schedule shape, and here's where it gets fun. simple/linear match BasicScheduler exactly, so it doubles as a baseline you can trust. beta and beta57 are beta-distribution spacings (beta57 skews toward the detail end - it's Boyo's own take on the "beta57" family you see in RES4LYF workflows, so it behaves like a cousin, not a clone). karras and exponential are the familiar curves forced into [0, 1] space. And then it goes off the rails in a good way: chaotic_logistic, chaotic_henon, and non_monotonic use seeded chaos-map spacing with reinjection events. Yes, really - and yes, they're mostly for seeing what happens.
• seed - for the chaotic and non-monotonic strategies. Ignored by the deterministic ones.

Outputs are the two things you'll wire:

• sigmas - the schedule, straight into SamplerCustomAdvanced.
• plot - an IMAGE of the schedule with every step value labelled, into a PreviewImage node. Running it once to eyeball the curve before committing is the whole point of the plot output.

Installing it

Same pack, same install:

cd ComfyUI/custom_nodes
git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes

or search Boyonodes in ComfyUI Manager, then restart. The plot is drawn with PIL - no matplotlib needed - and the beta strategies use scipy, both already in any normal ComfyUI install.

Where people get burned

• Sigmas stay in [0, 1] here, but your sampler must agree. This node is for flow models; feeding its output into a non-flow setup expecting [0, ~14] schedules will produce nonsense. The node even raises if you push sigma_min above 1.0.
• The chaotic strategies are a science experiment, not a fix. If you just want a good schedule for a distilled Wan or LTX model, simple or beta57 is the sane choice; treat chaotic_henon as a curiosity to show off, not a daily driver.
• Usual pack gotcha: if Boyonodes as a whole is missing from your node list, one of its audio/video modules failed to import - read the console traceback, install the named package, restart.

For anyone sampling flow models and wondering why a scheduler feels wrong, this node turns the black box into a line you can actually look at. That alone is worth the install.