Adept Scheduler (AkashicAOS Alt)

If you've read the AkashicAOS article, you know the plot: a schedule tuned for EQ-VAE SDXL models, a niche that almost nobody else targets. This is the "Alt" take on it - same target, different tuning philosophy, and it's the one to reach for when plain AkashicAOS feels like it's under-committing to detail.

The differences are specific and worth understanding. Where AkashicAOS uses a 0.85-power detail bias with a sinusoidal mid-boost, the Alt version pushes the bias harder - 0.78 power, which shifts even more step density into the low-sigma detail phase - and swaps the mid-boost for a shifted tanh crossover at t=0.55. The crossover point matters: the author's model of EQ-VAE is that its information-gain peaks later in the trajectory than a normal VAE's, so the step-dense region is parked at 0.55 instead of dead center. The third change is step-count-adaptive rho: at low step counts the schedule raises rho (up to 11) to concentrate on the detail phase where few steps need to count most, and it relaxes toward standard at high counts so extra steps stay meaningful instead of piling into an already-saturated region.

Practically, that means: if you're running EQ-VAE at low step counts (say 10–15), Alt is often the better sibling - the adaptive rho was built for exactly that situation. At 20–30 steps the difference between the two narrows, and it becomes a flavor choice between the sinusoidal mid-boost and the tanh crossover.

Inputs are the bare minimum again: model and steps (default 20), outputting SIGMAS into SamplerCustom's sigmas input. Like its sibling it's designed to feed the AkashicSolver v2 sampler, and the README's EQ-VAE recipe (tau 0.5–0.6, order 2, external rescaleCFG at 0.7) applies unchanged - just swap this node in for AkashicAOS if you want the stronger bias.

Install is the pack-wide one, seconds long, no hidden dependencies:

cd ComfyUI/custom_nodes
git clone https://github.com/nawka12/ComfyUI-Adept-Sampler

Restart, or install via ComfyUI Manager by searching "ComfyUI-Adept-Sampler".

The honest take: AkashicAOS Alt is a refinement, not a revolution - if the base schedule wasn't working for you, this is a slightly different flavor of the same idea rather than a different dish. But for low-step EQ-VAE work the adaptive rho genuinely does what it claims, and "the one with the stronger detail bias" is a useful thing to have in your back pocket when your EQ-VAE output reads soft.