Adept Scheduler (Advanced)

This pack ships over a dozen individual scheduler nodes, and this one is the shortcut: a single node with a dropdown that gives you access to all 18 schedules at once. It's the node to reach for when you're A/B testing schedules - instead of dragging a new node in and out of the graph for every experiment, you change one dropdown value and re-queue. Same seed, same sampler, different schedule, instant comparison.

The scheduler dropdown includes AOS-V, AOS-ε, AkashicAOS, AkashicAOS Alt, AkashicEQFlow, Entropic, SNR-Optimized, Constant-Rate, Adaptive-Optimized, Cosine-Annealed, LogSNR-Uniform, Tanh Mid-Boost, Exponential Tail, Jittered-Karras, Stochastic, JYS (Dynamic), Hybrid JYS-Karras, and AYS-SDXL. That's the whole menu from the pack's dedicated nodes, one knob.

Inputs:

• model - the loaded checkpoint. It's used to read the model's own sigma_min/sigma_max so the schedule is built for that checkpoint's noise range, not a hardcoded one.
• steps (1–10000, default 20) - how many sigma steps to generate.
• scheduler - the dropdown above.
• entropic_power (optional, default 6) - only consulted when the dropdown is on Entropic. It does nothing for the other 17, so don't be confused when changing it appears to do nothing.

Output is a SIGMAS socket, wired into SamplerCustom's sigmas input, with a sampler node alongside it. If you're using SamplerCustom, the pattern is [Load Checkpoint] → [Adept Scheduler (Advanced)] → [SamplerCustom].

The honest reason this node exists: most of those 18 schedules are niche. AOS-V and AOS-ε are the anime-optimized pair (v-prediction and epsilon respectively), the Akashic trio is for EQ-VAE models, AYS-SDXL is the research-backed default, JYS and Hybrid JYS-Karras are the low-step specialists, and the rest are experimental fills like Exponential Tail and Tanh Mid-Boost. You'll realistically cycle through five or six of them. But having them behind one dropdown means you can sweep the whole menu in a single session, which is genuinely useful for finding what a new model likes.

Install is the pack-wide one and takes seconds - pure Python, no requirements.txt, no model files:

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

Restart ComfyUI, or install via ComfyUI Manager by searching "ComfyUI-Adept-Sampler". It lands in sampling/adept/schedulers.

One workflow tip: when you find a schedule you like via this node, swap in the dedicated single-schedule node for that schedule in your final workflow. It makes the workflow self-documenting - someone reading your JSON knows exactly which schedule it's on, rather than trusting a dropdown value. If you're on SDXL and just want a solid default, AYS-SDXL is the one the research crowd would pick; if you're on a v-prediction model, AOS-V. Everything else is exploration.