Get Audio From Video

The audio from that video is a first-class asset now

The local video pipeline quietly became an audio pipeline somewhere along the way. LTX-2 generates audio and video together, Bernini S2V drives lipsync from an audio track, and dubbing workflows swap a voice track in and out. All of those need one thing first: the audio pulled out of a video as a standalone AUDIO object. That's this node - one input, one output, no widgets.

It's a small node, but it's the hinge for a whole category of workflows: re-using a soundtrack, feeding dialogue into a TTS replacement chain, or grabbing the audio from a generated clip so you can feed it back into a model that conditions on it.

How it works

Two-path extraction, in a sensible order:

1. FFmpeg first. The node tries to pull the audio track with FFmpeg, which is fast, lossless, and keeps the original sample rate and channels.
2. ComfyUI fallback. If FFmpeg isn't available or fails, it falls back to reading the audio straight out of the VIDEO's components (the waveform + sample rate the video object carries internally).

If the video genuinely has no audio track, it raises a clear error rather than returning silence. That's the one real failure mode, and it's the honest one.

Inputs and outputs

• video (required) - any VIDEO input: a loaded file, a generated clip from a decode node, an LTX output. If it has a soundtrack, it can be pulled.
• AUDIO out - a standard AUDIO object with waveform and sample_rate, so it plugs into anything else in the pack's audio world: easy audioMerge to mix it, easy makeAudioList to batch it, an audio VAE encode for LTX-2, or a Bernini S2V conditioning node that wants speech audio.

Setup and the one thing to remember

Install is the pack install - ComfyUI Manager → ComfyUI-Easy-Media, or git clone https://github.com/yolain/ComfyUI-Easy-Media.git into custom_nodes and restart. FFmpeg on your PATH is strongly recommended by the pack's README, and it's the primary extraction path here, so make sure ffmpeg -version works before you rely on it.

Where people get stuck:

• "The input VIDEO does not contain an audio track." - That's the node telling you the truth. Some generators output silent video (Bernini is the famous mute one - it has no audio at all), and no node can extract what isn't there.
• Silence even though it ran - check what you're feeding. If the upstream video was already silent, the fallback path can produce a zero-length waveform that some downstream nodes treat as "valid." Not a bug in the extractor; the source was empty.
• FFmpeg warnings scrolling by - normal. Extraction is quick but chatty on some containers.

It's not a glamorous node, and it shouldn't be. But if you're doing anything with the sound in your videos - dubbing, remixing, conditioning - this is the clean way in.