Send Audio

Most of the ComfyUI-Jumper pack is about pictures and prompts, but audio exists in ComfyUI too - and if you're doing anything with synchronized sound (audio-to-video models, voice tracks on generated clips, conditioning on a waveform), you might need to move that audio between machines. Send Audio is the node that does it: it takes a ComfyUI AUDIO value and throws it down the same socket the image and string nodes use.

How it works

Send Audio is the audio-flavored cousin of Send Images, minus the format menu. A ComfyUI AUDIO value is really a dict holding a waveform tensor and a sample_rate. The node serializes it into bytes - a 20-byte header carrying the sample rate, then the waveform saved through safetensors - wraps it in the pack's standard payload (size prefix, an audio format tag, SHA-256 hash, base64), and pushes it over TCP to a Receive Audio on the other machine. The receiver parses the sample-rate header, loads the tensor back, and rebuilds the AUDIO dict for you.

addr and mode are the pack's usual pair: addr defaults to 127.0.0.1:8282, and mode is Client (dial) or Server (listen). Match the opposite role on the Receive Audio end, and per the README run the whole thing through an SSH tunnel if the two machines aren't on the same network - plain TCP carries no encryption of its own.

The inputs that matter

• addr - host:port of the other side, default 127.0.0.1:8282.
• mode - Client or Server, opposite of the receiving node.
• audio - the AUDIO value to send. Wire it from an audio loader, a VAE decoder with an audio head, or whatever upstream node produced your waveform.
• order - chain input, default 0.

The only output is order (INT), which you pass to the next Jumper node in the chain - same rule as every node in this pack: one chain per workflow, matching chains on both sides, start at 0 with a Primitive Int and anchor the end in a PreviewAny. The counter is what keeps a Send Audio and a Receive Audio pointing at each other instead of crossing wires with a Send String on the same chain.

One thing Send Audio does not do is transcode. You get a lossless pass of the waveform tensor, sample rate intact - which is what you want when the receiving side is going to condition generation on that audio. Don't expect format conversion to MP3 or anything; it's a transport, not an audio editor.

Installing it

The pack has no dependencies beyond what ComfyUI already provides - empty requirements.txt, no model downloads. Install via ComfyUI Manager (search ComfyUI-Jumper) or:

cd ComfyUI/custom_nodes/
git clone https://github.com/myonmu0/ComfyUI-Jumper

Then restart ComfyUI.

Where people get burned

The pack's known issue applies to audio too: the cancel button won't stop a running Jumper node, so restart ComfyUI to abort a transfer. And the usual pairing gotcha - if the receiving side expects an image or a string and your Send Audio shows up, you'll get Receive unexpected data. because the format tag is checked on every message. Keep the chain order aligned, keep the tunnel up, and the waveform arrives intact.