ExponentialScheduler_motorway_edition

A scheduler is what decides how much noise to remove at each step - the shape of the denoise trajectory. ExponentialScheduler spreads those steps out exponentially between a max and min sigma, which is the aggressive-shape end of the scheduler spectrum. The _motorway_edition version from agilly1989/ComfyUI_agilly1989_motorway wraps ComfyUI's core node so the sigmas it computes land on the Motorway pipe under a key, instead of coming out of a socket.

The inputs

• steps (default 20) - how many sigmas to generate.
• sigma_max (default 14.614642) - the starting noise level, i.e. the high end of the schedule.
• sigma_min (default 0.0291675) - the ending noise level, the "almost clean" floor.
• OUTPUT_SIGMAS_key (default SIGMAS) - the key the sigma tensor is stored under on the Motorway.

Output is the MOTORWAY pipe; a motorway-ed KSampler/custom sampler reads the sigmas back via its INPUT_sigmas_key.

The defaults aren't random - 14.61 and 0.029 are the standard model noise floor/ceiling that most schedules anchor to. Exponential's character is in the middle: it front-loads denoising effort toward the loud end, which is great for getting composition in place early on curved DDPM-style trajectories.

Where it bites

Here's the part worth actually remembering. The community's sampler sweeps keep landing on the same verdict for modern models: Karras and exponential are universal failures on flow-matching architectures like Flux-derivative and Z-Image-family checkpoints. Those models want a balanced sigma schedule; exponential reshapes the curve aggressively, and on a near-straight flow trajectory that's a distortion, not a correction. So:

• SD 1.5 / SDXL (DDPM-style): exponential is a legitimate tool for low-step speed runs.
• Flux, SD3, and anything flow-based: pick a linear/normal/beta-style schedule instead. On those models this node is mostly a trap.

That's a grounded opinion, not a rule from the pack - the pack itself just wraps the core scheduler faithfully.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/agilly1989/ComfyUI_agilly1989_motorway

Restart ComfyUI, or install via ComfyUI Manager under "ComfyUI_agilly1989_motorway." No extra dependencies.

And the standing pack caveat: v1.1.7's pyproject.toml calls itself "HOTFIX - REMOVED MONKEYPATCH AND GENERATED NODES," and the clone nodes are commented out of __init__.py. Core Motorway nodes work; the _motorway_edition clones may not be in your menu until the author re-enables the generator. If it's missing, core ExponentialScheduler does the same math, and a Motorway 1x0 ramp gets the sigmas onto the bus manually.