LongCat Avatar Audio Crop

LongCat_Video_SM_AudioCrop is the pack's quiet utility: it takes an AUDIO input and cuts it down to the span between a start time and an end time. That's the whole job. You use it to strip silence off the front of a clip, to isolate the one sentence you actually want the avatar to speak, or to trim a long podcast segment down to the part worth animating. Unglamorous, but it sits directly upstream of the audio-encode node, and it saves you from paying generation time on dead air - which, given how slow LongCat sampling is, is real money.

How it works

Simple and predictable: you give it a start and end time as M:SS strings (defaults 0:00 to 1:00), it slices the audio payload in place, and returns the cropped clip as the same AUDIO type. It's an output node, so it can also show an inline preview of the cropped audio - handy when you're dialing in exactly where the voice starts. There's even a Crop Preview-style workflow where you can queue just this node to audition a crop without dragging the sampler and the whole video pipeline along. The crop stays on ComfyUI's native audio path, so whatever downstream nodes you hang off it get a normal AUDIO.

The inputs

Only three, and only one is one you'll change:

• audio - any ComfyUI AUDIO source (LoadAudio, or the output of LongCat Avatar Vocal Extract).
• start_time - string like 0:02, default 0:00.
• end_time - string like 0:14, default 1:00.

Output is the cropped audio (AUDIO), ready to feed LongCat Avatar Audio Encode.

Install

This node ships in the ComfyUI-LongCat-Avatar pack with all eight others:

cd ComfyUI/custom_nodes
git clone https://github.com/rookiestar28/ComfyUI-LongCat-Avatar
cd ComfyUI-LongCat-Avatar
pip install -r requirements.txt

or ComfyUI Manager → search ComfyUI-LongCat-Avatar → restart. NVIDIA + CUDA required.

Where people get burned

The format trap: times are strings in minutes:seconds, not raw seconds - 90 won't parse as 1:30, so write 1:30. If you crop past the end of the file you'll get whatever exists, so it's worth previewing before committing a long render. And keep the crop before the audio encode node in the graph: cropping after encoding is pointless, since the conditioning was built from the untrimmed audio and the sampler will just cut its output to the source length anyway. Trim early, save yourself a confused hour of "why is the avatar silent for the first second."