Tan Scheduler 2 Simple

The scheduler is the half of the sampling equation that decides how much noise comes off at each step, and Tan Scheduler 2 Simple lets you draw that curve yourself using a tangent-based shape with two adjustable pivots. If the stock schedulers (beta, simple, Karras) don't give you the noise distribution you want, this is a build-your-own.

Worth being honest up front: this is a power-user node. On a flow-matching model - Flux, Chroma, SD3.5, Z-Image, Anima, the whole family RES4LYF targets - the community's hard-won lesson is that these models dislike aggressive reshaping of the sigma curve. A big 62-sampler sweep on Z-Image found Karras and exponential failing across the board precisely because they redistribute denoising effort too hard, while balanced schedules like beta and simple held up. So a fully custom curve is a sharp tool that can cut you. It shines when you know exactly what you're doing; for most people beta57 or plain simple is the smarter default.

How it works

It samples a tangent-shaped function across your step count, anchored by start, middle and end sigma values, with two pivot/slope pairs bending the curve. The result is a SIGMAS schedule - the same object a stock scheduler node outputs - that you route into a sampler.

The inputs and outputs that matter

• steps (default 40) - number of steps in the schedule.
• start / middle / end (defaults 1 / 0.5 / 0) - the sigma anchors the curve passes through, normalized. end at 0 means it finishes fully denoised.
• pivot_1 / pivot_2 and slope_1 / slope_2 - where the tangent bends and how sharply. These are the shape controls; nudge one at a time.
• pad (default false) - whether to pad the schedule.

Output is sigmas (SIGMAS) - feed it to a RES4LYF sampler (or any custom-sampling node that takes sigmas).

How to install it

Ships with RES4LYF. ComfyUI Manager: search RES4LYF, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt

Portable builds use the embedded pip. Restart, hard-refresh (F5).

Common issues

The first failure mode is producing a curve that isn't monotonic - if your pivots and slopes make the sigmas wander up and down where they should be steadily falling, the sampler gets a schedule that makes no physical sense and the image falls apart. Keep the curve heading downhill from start to end. Second, remember the flow-matching caveat above: if you're on Flux or Z-Image and your custom curve looks worse than the boring simple scheduler, that's expected, not a bug - these models reward restraint. And a classic RES4LYF scheduler trap to keep in mind across the pack: some schedulers apply their own internal sigma shift and ignore the workflow's, so if a shift node elsewhere in your graph seems to do nothing, suspect the scheduler before you suspect the shift.