Audio Metadata

Video workflows keep colliding with a question the graph can't answer: how long is this audio? You loaded a track to score a render and you want the frame count to come out exactly the length of the music. Or you're about to encode a video whose audio track might be silent - and you'd rather know before you burn the encode. Audio Metadata is the node that unpacks an AUDIO value into plain numbers.

Under WAS Suite/IO, it takes an audio input (from Load Video (Advanced), Load Video (Upload), or any node with an AUDIO output) and reads the bundle the way ComfyUI stores sound: a waveform tensor plus its sample rate. What it hands back is a full instrument panel:

• duration - seconds one clip plays (samples ÷ sample rate). Feed it to a frame count and a render comes out the length of the music. This is the headline output.
• sample_rate - samples per second: 44100 for CD audio, 48000 for most video. Compare two tracks before mixing clips that won't line up.
• channels - mono (1) or stereo (2).
• samples - raw sample count per channel. At 48000 Hz, 480000 samples is ten seconds.
• batch_size - how many clips are stacked in this AUDIO. A loader answers 1; more means something upstream stacked several.
• is_silent - true when the loudest sample is at or below your silence_threshold, and also true when nothing readable arrived at all.
• peak - the loudest single sample anywhere, as distance from zero: 0.0 is silence, 1.0 is full scale, above 1.0 means it will clip when written out.
• rms - the average level on the same 0.0–1.0 scale. RMS is what sounds loud: mastered music sits near 0.2, a quiet dialogue take nearer 0.02, so a pair of tracks with similar peaks can still read very differently here.
• summary - the whole reading as one line (44100 Hz, 2 ch, 132300 samples, 3.000 s, batch 1, peak 0.813, rms 0.204), with , silent appended when it is, or no audio when nothing readable arrived. The same text is drawn on the node itself, since it's an output node.

The one setting, silence_threshold (0.0–1.0, default 0.0), decides what counts as silence. 0.0 means only a track silent to the last bit counts; nudge it to 0.001 (about −60 dBFS) and it also catches a noise floor or a tail of dither. Sample values run 0.0 to 1.0, so think of this as a small decimal, not a percentage.

The design choice that makes it useful

A load that found no sound doesn't error - it answers zeroes and is_silent true, and keeps running. That's deliberate, and it's what makes this a branch node rather than a crash point: wire is_silent into a switch or a gate and a graph can skip an encode that would only write a silent track, or route in a music bed when the source turned out mute. And duration wired into a frame-count calculation is the whole "video matches the music" trick - no more guessing frames, re-rendering, discovering the beat lands in the wrong place.

The peak output doubles as a loudness health check: if peak reads above 1.0, the track will clip when it's written, and catching that at this node beats discovering it in the finished file.

Installing

Part of WAS Node Suite v3 (WASasquatch/was-node-suite-comfyui), WASasquatch's MIT pack - going since 2023, over a million downloads, and one of the packs that's been in the ecosystem's essentials conversation from the start. Audio metadata reads the AUDIO value ComfyUI already holds - no extra packages, no models, nothing downloaded. Requires ComfyUI 0.14.0+ (the v3 suite targets ComfyUI's newer node backend).

Via ComfyUI Manager, search WAS Node Suite v3, or clone:

cd ComfyUI/custom_nodes
git clone https://github.com/WASasquatch/was-node-suite-comfyui

Restart ComfyUI and it appears under WAS Suite/IO. If a video load answers nothing on its AUDIO socket, that usually means the file has no sound track at all - which this node will happily report as silence rather than an error. If the node itself is missing after install, update ComfyUI first.