ElevenLabs Pro - Audio Trim

AI audio comes with dead air. TTS clips pad their starts, STS re-renders add a beat of silence, and isolation runs often leave you a second of nothing on the front. ElevenLabsPro_AudioTrim is the local, free node that cuts any AUDIO clip to a start and end point, so you can clean up before export instead of fixing it in an editor afterward.

Inputs are exactly three: audio, start_seconds, and end_seconds. The one that trips people up is end_seconds - a value of 0 means "trim to the end of the clip," not "trim to zero length." That's the tooltip's way of giving you a "to the end" default, and it's the behavior you'll use nine times out of ten: set start_seconds to shave the leading silence, leave end_seconds at 0, done. Set both and you get a precise window, down to hundredths of a second. The sliders cap at 3600 seconds (an hour), and if your start lands at or past the end, the node returns a sliver of silence rather than crashing - it's defensive, so a bad range degrades gracefully into a near-empty clip instead of a stack trace.

Mechanically it converts your seconds to sample indices against the clip's sample rate, slices the waveform tensor, and returns a new AUDIO dict at the same rate. Pure torch, no API call, no key, no credits. One output, audio.

It slots neatly into the pack's utility chain. Trim the dead air off each generated clip, Normalize them all to the same loudness, then Concat them into a single take - that's the canonical three-node assembly pipeline, and Trim is the first step because it shrinks what everything after it has to chew on. It's also the right pre-step before an API call: Audio Isolation and Speech to Text bill or process based on what you send, so trimming 5 seconds of room tone off the front of a 3-minute clip before you transcribe is pure efficiency.

Install is the pack standard:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-ElevenLabs-Pro.git
pip install -r ComfyUI-ElevenLabs-Pro/requirements.txt

Restart ComfyUI, or install "ComfyUI-ElevenLabs-Pro" via ComfyUI Manager. Requirements are just requests and soundfile; this node only needs torch.

Two gotchas worth knowing. First, this is sample-accurate slicing, not a silence detector - it cuts at the seconds you give it, it won't find or remove the dead air itself. If you don't know where the silence starts, check the Metadata node's duration_seconds or just eyeball a preview. Second, unlike the paid nodes, it has no caching concern at all: it's deterministic and instant, so you can re-queue freely and it costs nothing. If you need a trim that adapts to content (cut all leading silence regardless of how much there is), this bare-bones node isn't that - but for the 95% case of "shave the fixed delay off my clip," it's exactly right.