Demux · Silent Video

The other half of the demux pair. Where Demux · Audio Track hands you the sound from a video, this node gives you the video with the sound removed - a video output that's the same footage, now silent. One input (video), one output (video), and nothing else to configure.

Why strip audio off a clip you still want to use? Because "video with its original audio" is often not what you want downstream. Maybe you're about to lay down new music and don't want the old track bleeding through. Maybe you processed the audio separately (repair, stem split, a fresh mix) and want to re-attach the cleaned version. Either way, the silent video is the clean slate - and the pack's demux toolbar action spawns both halves at once, so you can route picture and sound down separate pipelines and recombine them when you're done.

Mechanically it's a container-level demux, not a re-encode: the audio stream is dropped and the video stream passes through essentially untouched. That means it's instant, lossless, CPU-only, and the video quality is identical to the source. No models, no GPU, no trade-offs - the only "cost" is remembering that the output really is silent, so don't panic when you play it back and hear nothing. That's the feature.

The common beginner stumble is wiring the original video downstream by mistake after you've already demuxed - you get your old audio back and wonder why your new mix never attached. If that happens, trace the wire: the silent output is the video socket of this node, and re-adding audio is a separate mux/volume stage's job, not this node's. The pair is meant to be split-then-recombined, and once you've done it once, it clicks and you'll reach for it every time a clip comes with sound you don't want.

Install is the pack-wide story: ComfyUI Manager search "ComfyTV", or git clone https://github.com/jtydhr88/ComfyTV into custom_nodes/, then a full backend restart (Desktop/macOS: clone by absolute path into the running instance or the pack won't register). Zero extra pip dependencies - PyAV stream handling, no models, CPU-only.