FL Audio Crop

Most audio-driven video work starts with the same annoyance: you've got a three-minute song loaded, but you only want to animate to the chorus that lands at 1:12. FL_Audio_Crop is the little scissors for that. Hand it an AUDIO input and a start and end time, and it gives you back just that slice - no re-encoding in a separate app, no math, done inside the graph.

It's not glamorous, but it's the node that makes the rest of Fill-Nodes' audio suite usable. Beat analyzers, drum detectors, frame calculators, reactive effects - they all work off whatever audio you feed them, and you almost never want the whole file. Crop first, then analyze, then generate. Cutting to the segment you care about also means the downstream frame budget stays sane instead of asking your video model for four minutes of frames.

How it works

The node takes the waveform, converts your start and end times to sample positions, and returns the samples in between. Times can be written either as MM:SS (like 1:12) or as plain seconds (like 72) - whichever is easier. If you ask for an end time past the actual length of the clip, or a negative start, it clamps to the real boundaries rather than erroring, so you can be a little sloppy and still get a valid result.

The inputs and outputs that matter

Three inputs, all required:

• audio (AUDIO) - the track to cut. Usually straight off a Load Audio node, or the audio pulled out of a video.
• start_time (STRING, default "0:00") - where the slice begins. MM:SS or seconds.
• end_time (STRING, default "1:00") - where it ends. Same formats.

Output is a single audio (AUDIO), the cropped waveform, ready to plug into a BPM analyzer, a frame calculator, a lip-sync node, or straight into a video-gen node that takes audio.

When you'd reach for it

Any time the audio you have is longer than the moment you want to work with. Isolating a chorus for a music-video shot. Grabbing a single spoken line for a talking-head lip-sync. Trimming silence off the head of a recording before beat detection so the beats line up. It pairs naturally with FL_AudioFrameCalculator - crop to your segment, then ask how many frames at your FPS it takes to cover it.

Installing it

ComfyUI Manager → search ComfyUI_Fill-Nodes → install → restart. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_Fill-Nodes

then restart. The audio nodes in this pack lean on standard audio libraries (librosa/torchaudio), which the pack installs on first launch - Fill-Nodes is a big collection, so expect a chunky dependency pull the first time.

Common issues & troubleshooting

Format matters. Times are strings, not numbers. 1:12 is one minute twelve; 72 is the same in seconds. 72.0 as an intended timestamp is 72 seconds, not 72 minutes - don't overthink it, but do keep MM:SS and raw-seconds straight in your head.

End before start gives you nothing useful. If end_time is earlier than start_time, you're asking for an empty or backwards slice. Double-check the order.

No audio input, no output. It needs a real AUDIO tensor upstream. If you're pulling audio out of a video, make sure that extraction actually produced a track before wiring it in.