Legacy2_ClownsharKSamplerAutomation_Advanced

If you loaded an old RES4LYF workflow and found a node called Legacy2_ClownsharKSamplerAutomation_Advanced in your graph, here's why it's still there: the pack's README says it straight, "the previous versions will remain available but with 'Legacy' prepended to their names." RES4LYF got rewritten into a cleaner all-in-one ClownsharKSampler, but rather than break every workflow that predates it, the old pieces got kept alive under this prefix. This is one of those pieces. If you're starting a workflow fresh today, you almost certainly want the current sampler's built-in fields instead - this node exists to keep old graphs loading, not to be your first choice.

What it actually does

It doesn't sample anything itself. It's a pure bundler: you feed it a handful of separate schedule curves and it packages them into a single AUTOMATION object that plugs into the legacy ClownsharKSampler's automation slot. This is the concrete form of something RES4LYF cares about a lot - modulating sampler parameters over time instead of leaving them as one fixed number for the whole run. Rather than setting eta to 0.25 and leaving it there for every step, you can feed in a curve so it changes step by step.

Every input is a SIGMAS-typed slot, meaning each one expects the output of an actual schedule-building node elsewhere in your graph - not a plain number.

The inputs and outputs that matter

All optional, so you only wire in the schedules you actually want to vary:

• etas - a step-varying schedule for the eta value that controls how much SDE noise gets added after each step.
• etas_substep - the same idea but for the eta used inside the multi-substep samplers (the ones ending in 2s/3s/5s).
• s_noises - a schedule for the extra noise-scaling multiplier.
• unsample_resample_scales - a schedule specific to unsampling/resampling passes, the noise-inversion img2img technique RES4LYF's README covers at length.
• frame_weights and frame_weights_bkg - per-frame weighting for foreground and background, the video-side hook behind the pack's temporal conditioning feature (changing behavior frame by frame).
• automation - an existing AUTOMATION object to extend rather than build from scratch.

Output: automation (AUTOMATION), which feeds straight into the legacy ClownsharKSampler's automation input.

How to install it

Via ComfyUI Manager (search "RES4LYF"), or manually:

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

then restart.

Common issues & troubleshooting

Wiring the node in isn't enough - you need at least one schedule connected. Every input here is optional, which means it's entirely possible to drop this node into your graph and have it do nothing, because nothing's actually feeding the automation object. If a parameter you expected to vary over the run is staying flat, check that the corresponding SIGMAS input is actually connected.

Red link on connection = wrong node type upstream. Each of these wants the output of a genuine schedule node, not a float widget. If ComfyUI refuses the connection, you tried to feed it a raw number where it wants a curve.

It's the legacy branch, so it's frozen. New RES4LYF development goes into the current all-in-one sampler; this node won't gain new capability. If you're not maintaining an old workflow, look at the current ClownsharKSampler's documentation and example workflows instead - most of what this node does is exposed there directly, without a separate automation object to assemble.