ElevenLabs Pro - Audio Concat

Once you've generated five paragraphs of TTS narration, you have five separate clips and one continuous voiceover to build. ElevenLabsPro_AudioConcat is the local, free node that merges up to eight AUDIO inputs into a single clip along the time axis - and it's smart about the messes you'd otherwise have to fix by hand.

The smart part is how it handles mismatches, because it will encounter them constantly. Sample rates differ between clips? It resamples everything to the first input's rate, using torchaudio when it's installed and a linear-interpolation fallback when it isn't. Channel counts differ? It finds the widest clip and broadcasts mono up to match it. Both of those happen automatically, which is the difference between this node and a naive torch.cat someone threw together. There's also a gap_seconds slider (0–30, in 0.05 steps) that inserts a silence buffer between each clip - exactly what you want between narration paragraphs, or when you're spacing out sound effects in a library build.

Inputs are straightforward: audio1 through audio8, all optional, and gap_seconds. Any slot you leave empty is simply skipped, so you can chain partial setups without rewiring. One output, audio, ready for Save Audio, a preview node, or more downstream processing.

Mechanically it normalizes every incoming waveform to [1, C, T], resamples if needed, broadcasts channels, and concatenates along the time dimension - with the silence gaps inserted as zeros at the target sample rate. Pure Python plus torch, no network, no API key, no credits. It'll raise a clear error if you feed it nothing at all.

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. The pack's declared dependencies are just requests and soundfile - but note the resampler: if you want the high-quality torchaudio path (rather than the linear-interp fallback) you'll need torchaudio in your environment. Most ComfyUI installs have it; if your clips come out sounding slightly soft on pitch when rates mismatch, check that.

Where people get burned: order matters, because the output inherits the first clip's sample rate. If your first clip is a low-res 8 kHz capture and the rest are 44.1 kHz TTS, everything gets dragged down to 8 kHz. Feed the highest-quality clip into audio1. Also, this node concatenates - it does not mix, crossfade, or duck. If you want overlapping layers or a music-under-voice bed, this isn't the tool; it's for linear assembly, which is exactly the job it nails. And if you're building a long audiobook, remember the 30-second max gap and that gap silence counts toward the final duration - the Metadata node will show you exactly what you produced.