The Golden Scheduler

The namesake, and it's not a gimmick

This is the node the whole pack is named after, and it's the one Extraltodeus demonstrably put the most work into - there are side-by-side renders in the README comparing it against Karras on his own model. The idea is simple: instead of using Karras's 1/7 exponent or a plain linear blend, The Golden Scheduler uses the golden ratio (φ ≈ 1.618) as the exponent that bends the curve between the model's sigmax and sigmin. For each step x out of steps:

(1 - x/(steps-1))**phi * sigmax + (x/(steps-1))**phi * sigmin

It's the exact same shape family as the Karras formula - both are weighted blends between the two extremes - just with a friendlier exponent. The result reads as a middle ground between simple and Karras: a smooth schedule that spends more time at the noisy end and eases off, without Karras's aggressive middle push. The author recommends it with dpmpp2m, euler, and lms, which lines up with where the "gentle curve" schedulers historically shine on SD-class models.

The input side is minimal: model, steps (default 20), and the same sgm boolean the Manual scheduler has. Flip sgm and it adds then drops a step to align to the trailing SGM style used by distilled checkpoints. One SIGMAS output, wired into a KSamplerAdvanced's sigmas input like any external scheduler.

The honest take

On SD 1.5 and SDXL finetunes, where DPM++ 2M Karras is still the community default, this is a pleasant alternative when you want something gentler than Karras without dropping to a flat linear schedule. Try it side by side with Karras at 20–30 steps on a fixed seed; you'll likely see softer contrast and slightly different composition - exactly what a less aggressive curve does.

But keep the architecture caveat in mind: schedule reshaping is a DDPM-era trick. On flow-matching models (Flux, Z-Image, Klein, Anima) the community has repeatedly found that Karras-style curves actively hurt - the straighter the denoising trajectory, the more a reshaped sigma curve costs you. The Golden Scheduler is squarely in that same family, so don't reach for it on a flow-matching checkpoint; Euler on a conservative schedule is still the right tool there.

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. If the pack fails to import, it's the classic dependency gap: requirements.txt lists only asteval but the code loads matplotlib and scipy, so a minimal venv may need pip install matplotlib scipy asteval.

It's a niche pack - a couple of dozen community mentions around 2024–25, mostly in its first year - so you won't see this in mainstream workflow shares. That's fine. It's a tinkerer's tool, and the tinkering is the point.