Sigma Schedule Interp. Mean 🎭🅐🅓

Displayed as Sigma Schedule Interp. Mean, this is the "Interp." version of ADE_SigmaScheduleWeightedAverage. Where that node blends two schedules at one fixed ratio for the whole run, this one lets the ratio itself shift from a start value to an end value across the schedule.

A pattern worth recognizing

A plain node, and an "Interp." sibling that ramps between two endpoints instead of holding one static value - this shows up repeatedly across this pack. LoRA Hook Keyframes has an interpolating variant, the PE Adjust family has ramping options, ContextRef keyframes do too. Once you've spotted the pattern here, you'll recognize it elsewhere in the pack faster.

Practically: ADE_SigmaScheduleWeightedAverage picks one weight_A and holds it. This node instead takes weight_A_Start and weight_A_End, and ramps between them across the schedule using whatever curve interpolation selects - so you can favor schedule A early in sampling and shift toward schedule B later (or vice versa), rather than a fixed mix throughout.

One default worth noticing: weight_A_Start and weight_A_End both default to 0.5. Left untouched, that's functionally identical to the plain Weighted Mean node at 0.5 - a flat blend, no ramp. You have to actually separate the two values to get any ramping behavior; the interpolation curve has nothing to interpolate between if both endpoints are the same.

Inputs and output

All required:

• schedule_A, schedule_B (SIGMA_SCHEDULE) - the two schedules to blend.
• weight_A_Start (default 0.5, range 0–1, step 0.001) - the mix ratio at the start of the schedule.
• weight_A_End (default 0.5, range 0–1, step 0.001) - the mix ratio at the end.
• interpolation - the curve shape used to ramp between them.

Output is a single SIGMA_SCHEDULE, feeding ADE_SigmaScheduleToSigmas for a concrete SIGMAS list, or the sigma_schedule slot on Sample Settings.

How to install it

ComfyUI Manager: search AnimateDiff Evolved, confirm the author is Kosinkadink, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Kosinkadink/ComfyUI-AnimateDiff-Evolved

No model download - pure schedule math, typically fed by ADE_RawSigmaSchedule-built objects.

Common issues & troubleshooting

Set both weights and nothing seems to ramp. If weight_A_Start and weight_A_End are the same value, that's expected - a flat weight produces a static blend, identical to the plain Weighted Mean node. Separate the two values to get an actual ramp.

Not sure which interpolation option to pick. Our sources don't give strong guidance here beyond the dropdown itself - start with the default option and compare a couple of alternatives on a short, cheap test render rather than guessing on a long one.

Result is hard to distinguish from the plain Weighted Mean node. If your start and end weights are close together, that's expected - the ramp is subtle by design at small gaps. Widen the gap between weight_A_Start and weight_A_End if you want a more pronounced shift.

Wondering if you need this level of control. For nearly all AnimateDiff work, no - this whole sigma-schedule-combination corner is deep custom-sampler territory. Picking one named beta_schedule preset on your loader is the right call for the large majority of workflows; reach for this only once you have a specific, deliberate reason to shift a noise curve's blend over the course of sampling.