Clip Amplitude

Raw amplitude from BatchAmplitudeSchedule is spiky. One loud snare hit can dwarf the rest of the track, which means your animation twitches on the hits and ignores everything else. ClipAmplitude is the fix: it clamps every value into a range you choose, so anything above max_amplitude gets floored at the max, and anything below min_amplitude gets lifted up to the min.

Think of it as a soft compressor for your curve. Flooring the top kills the spikes; raising the floor stops the curve from collapsing to zero in the quiet bits. Do both and a track with one enormous transient becomes a curve that moves across its whole dynamic range - which is exactly what you want when the curve is driving a ControlNet weight or a denoise schedule.

The inputs

• max_amplitude (default 1000) - everything above this becomes this. The default suits raw max/sum amplitude output, which produces big numbers.
• min_amplitude (optional, default 0) - everything below this becomes this. Leave it at 0 for pure spike-cutting; raise it if your curve keeps going flat in quiet sections.

Output: amplitude - same length, same AMPLITUDE type, just clamped. That means you can chain it (clip, then normalize, then gate) and the pack's types stay intact all the way down.

Where people get burned

The default max of 1000 is calibrated for un-normalized amplitude. If you're clipping after NormalizeAmplitude - where values live in 0–1 - a max of 1000 does nothing at all. The pack's own example clips raw amplitude to a max of 560 and a min of 500, which shows the other classic move: a tight window like that isn't just compression, it's a way to isolate a specific intensity band so the curve only reacts to a certain loudness range. Also note min_amplitude can't exceed max_amplitude - flip them and the node raises a ValueError rather than guessing.

Install is standard for this pack: ComfyUI Manager, search ComfyUI-AudioScheduler, or clone https://github.com/a1lazydog/ComfyUI-AudioScheduler into custom_nodes and restart. No models, no VRAM, CPU-only.