Trim Audio Duration

Generated audio never starts where you want it to. Stable Audio and ACE-Step happily give you lead-in silence, trailing noise, or a first bar you don't want - and TrimAudioDuration is how you cut it down to the good part. It slices an AUDIO tensor into a chosen time range, which makes it the most constantly useful of the audio editing nodes: you'll reach for it before almost every concat, merge, or save.

It's part of the September 2025 batch of built-in audio editing nodes, and it's about as direct as a node gets. Two numbers in, a shorter track out.

How it works

The node works in seconds, not frames - which is the first thing to appreciate, because it makes trimming feel like a DAW rather than a tensor op. Internally it converts your seconds to sample frames using the track's own sample rate, then slices the waveform.

The two inputs that matter:

• start_index (FLOAT, default 0) - where to start, in seconds. It supports sub-second precision, and here's the genuinely nice bit: negative values count from the end of the track. So -10 starts the trim ten seconds before the track ends. That's a lifesaver for grabbing the tail of a generation without first finding out how long it is.
• duration (FLOAT, default 60) - how long the trimmed segment should be, in seconds. The end of the slice is simply start + duration, clamped to the track length.

The one restriction: the start has to land before the end. If your start plus duration overshoots the track, the node clamps to the actual end rather than erroring - so a trim of start=5, duration=100 on a 30-second track just gives you the last 25 seconds.

Inputs and output

• audio (AUDIO) - the track to slice.
• start_index (FLOAT) - start in seconds; negative counts from the end.
• duration (FLOAT) - length of the segment in seconds.
• AUDIO (output) - the sliced segment.

Getting it

Ships with ComfyUI core since September 2025. Nothing to install.

Common issues

The main thing that trips people is expecting timeline-style clips with "end" values instead of start-plus-duration. It's start and length, not start and end - set duration to the segment length you want, not the end timestamp. And if you trim right up to the edge of a section, remember this node does a hard cut with no fade; for speech that's fine, for music you may want to trim a hair shorter or longer to land on a quieter moment. After trimming, feed the result to AudioConcat to assemble clips, or to PreviewAudio to check the cut by ear.