Preview Audio

PreviewAudio is the audio equivalent of the Preview Image node: a dead-end that plays your AUDIO tensor in the browser so you can actually hear what your pipeline produced before you save it anywhere. If you've built any kind of audio workflow - Stable Audio generation, an ACE-Step track, a voiceover you layered with MusicGen output - this is how you audition it without cluttering your output folder.

The name undersells it slightly. "Preview" sounds like a nice-to-have, but in practice it's the node you want at the end of every experimental chain, because iterating on audio blind is miserable. Sticking PreviewAudio after your merge, EQ, or volume stage turns a workflow into something you can tune by ear in seconds.

How it works

It's a terminal (output) node: it takes the AUDIO input, hands it to the frontend for playback, and does zero processing itself. The audio plays inline in the node's UI - no files written to disk, nothing left behind in output/.

One thing worth knowing: as of mid-2026 the preview and save nodes all got an AUDIO output socket added for schema uniformity, so PreviewAudio also passes its input through. That means you can chain it mid-graph if you like, or keep it as a true terminal node. Most people leave it at the end and don't connect the output.

Inputs and output

• audio (AUDIO) - the track to audition. Comes from LoadAudio, RecordAudio, or any processing node (AudioConcat, AudioMerge, AudioEqualizer3Band, and so on).
• audio (AUDIO, output) - passthrough of the same track.

A workflow worth stealing: PreviewAudio in parallel with your save node. Wire the same AUDIO into both, keep the save disabled while you tune, and flip it on when the preview sounds right. That's the closest thing ComfyUI has to a DAW's monitor channel, and it keeps your output folder free of the twelve rejected takes.

Getting it

Ships with ComfyUI core, present since the audio support landed in mid-2024. Nothing to install, no models to download - it's purely a listener.

Common issues

The one real error you'll hit is the "input audio is None (source video may have no audio track)" complaint. It happens when whatever feeds the node produced no audio - typically a video source with a silent or missing audio track. Fix it upstream: check the source actually has audio (LoadAudio will happily decode a video's first track), or feed PreviewAudio from an audio-specific path.

Beyond that, it's about as failure-proof as ComfyUI gets. If you can't hear anything, check your system audio, then check that you're not expecting sound from a node upstream that genuinely output silence (this is what EmptyAudio gives you, and it's silent by design).