← ComfyTV Audio

ComfyTV is a walled garden with a very open door: its stages talk to each other in ComfyTV's own data types (COMFYTV_AUDIO, COMFYTV_VIDEO, COMFYTV_IMAGE), and the Bridge nodes are the door. BridgeFromAudio takes a ComfyTV audio track and hands it back to the rest of ComfyUI as a standard AUDIO output - the exact type any other audio custom node expects.

That's the whole node: one input (audio, a COMFYTV_AUDIO), one output (AUDIO). No parameters, no settings. It exists because ComfyTV's audio isn't a tensor you can just hand to a third-party node - it's a reference to a rendered file in the project store - so the bridge materializes it: it loads the WAV with torchaudio and hands the downstream node a real waveform + sample rate. Downstream, your favorite third-party audio effects, analyzers, or save nodes work exactly as they would on any other ComfyUI audio.

The complement is → ComfyTV Audio (BridgeToAudio), which goes the other direction: a normal ComfyUI audio node's output becomes a ComfyTV track. Together they're how you mix and match - run ComfyTV's stem split, hand the vocals through the bridge into a plugin you already trust, and bridge the result back into the ComfyTV project. The README's bridges guide frames it as connecting "any plugin" into a ComfyTV pipeline, and this is the audio half of that promise.

Honest framing: as a single node it's plumbing - you won't build a workflow around it, you'll drop it in when the wire colors don't match. The two things to remember: it needs a real audio input (it errors if you feed it an empty reference), and it materializes a file on disk, so there's a small round-trip cost per call rather than zero-copy tensor passing. That's fine for almost everything you'll do; for the rare case where you're chaining a dozen third-party audio nodes in a tight loop, batch your bridges rather than scattering one per node.

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). Zero extra pip dependencies - torchaudio ships with ComfyUI, no models.