ClownOptions Automation

Most sampler settings are a single number you set once. ClownOptions Automation lets you make them a curve instead - a value that changes from step to step across the sampling run. Want more SDE noise early and less late? A different eta at each step? This is the node that feeds those per-step schedules into a RES4LYF sampler. And for Wan video, it carries frame_weights, so you can vary the effect frame-by-frame.

This is the concrete expression of RES4LYF's founding idea. ClownsharkBatwing has said the whole point of the pack is that "modulating parameters vs. time" produces large gains in image quality - that the sampling process isn't one setting, it's a trajectory you can shape. Automation is where you actually do that shaping: instead of a flat eta, you hand the sampler a SIGMAS line that is the eta schedule, one value per step. It's a power feature, and like most of the pack it rewards experimentation over documentation.

How it works

Each input takes a SIGMAS-typed list of values - one per step - and the node packs them into an OPTIONS object the sampler reads. Where the sampler would normally use a constant, it now walks your list. You generate those lists with the pack's sigma/curve nodes and route them in here.

The inputs and outputs that matter

Everything here is optional - supply only the parameters you want to automate and leave the rest to the sampler's own single values.

• options (OPTIONS) out - into your sampler's options input; chain with other ClownOptions nodes via the optional options input.
• etas (SIGMAS) - a per-step schedule for eta, the SDE noise-injection amount. The most useful one to automate: front-load noise for variety, taper it for a clean finish.
• s_noises (SIGMAS) - per-step schedule for the SDE noise scale (s_noise).
• etas_substep / s_noises_substep - the same, but for solver substeps (only meaningful on 2s/3s/5s samplers).
• epsilon_scales (SIGMAS) - per-step scaling of the model's prediction; a finer, more experimental lever.
• frame_weights (SIGMAS) - for video: weight the effect per frame. This is the hook for Wan temporal work, letting a setting ramp across the clip.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manual:

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

then restart and hard-refresh (F5). Nothing to download.

Common issues & troubleshooting

Nothing changed. Two usual causes: you left every input empty (then it passes through unchanged), or the SIGMAS list you fed doesn't line up with your step count. A schedule shorter than your steps can leave the tail on defaults.

The result got noisy or unstable. You're probably automating eta/s_noise too high across too many steps. SDE noise stacks - a schedule that holds a big eta all the way through is a lot of injected noise. Taper it toward the end.

frame_weights did nothing on an image. It's a video lever - it only means something when you're sampling frames (Wan). On a single image there are no frames to weight.