ElevenLabs Pro - Audio Isolation

Your voice track has traffic hum, AC rumble, and a fan somewhere in the room. ElevenLabsPro_AudioIsolation is the node that scrubs it, turning a noisy recording into something that doesn't sound like it was captured in a laundromat. It's the pack's audio-cleanup API node: feed it AUDIO, get back the same content with the background stripped out.

Under the hood it's a single multipart POST to ElevenLabs' /v1/audio-isolation endpoint. The node converts your ComfyUI audio tensor into WAV bytes, sends them up with your API key, and decodes the response back into an AUDIO dict using the same output_format list every paid node shares. The isolation itself - separating vocals from noise, or music from vocals - happens entirely in ElevenLabs' cloud, so like every paid node here it costs credits and needs an API key. It's also cached by input hash, so re-queuing the same clip won't re-bill you; change the audio and you pay again.

Inputs are minimal: api_key, audio, and output_format. That's the whole surface. One output, audio, which you'd typically route to a preview, Save Audio, or straight into Speech to Speech for further processing.

There are two distinct jobs people use this for, and it's worth knowing which you're doing. Cleaning a voice recording for a podcast or voiceover - removing room tone and hum before you normalize and export - is the classic one, and it's genuinely good at it. The other is stem work: isolating vocals from a full mix. The same endpoint handles it, but don't expect a perfect karaoke split on dense music; it's a separator, not a magician. Either way, the output is what the API considers "isolated," and you don't get a knob to dial how much noise stays - this node is deliberately bare-bones. If you want a partial denoise, that's a different tool entirely.

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 - no local models to download, because the separation runs server-side.

Where people get burned: tier and format limits. If your output_format is mp3_44100_192 or one of the opus variants, you need Creator tier or better - on a free plan the call fails with a permission error. And the input file size counts against request limits; this is not the tool for hour-long recordings, so trim your clip to what actually needs cleaning first (Audio Trim is handy for that). One more workflow tip: isolate first, then transcribe or STS. Feeding a clean signal into Speech to Text gives you markedly better transcripts than sending the noisy original - the order genuinely matters.