H3 Audio

H3 generates audio natively - dialogue, sound effects, room tone, all in the same pass as the picture. That's the whole point of the model, and it means an audio reference is a first-class thing you can hand it, not a file it ignores. H3 Audio is the node that wraps a wired AUDIO into an H3 reference: an <Audio j> the prompt can name, chained onto the pack's h3_inputs wire.

How it works

It's about as thin as a node gets. One required input, audio, one optional h3_inputs to chain onto, one output h3_inputs to chain onward. The audio becomes an <Audio j> reference - numbered in wire order alongside any <Picture i> and <Video k> tags - and your prompt on H3 Encode refers to it by that tag, say "match the voice in <Audio 1>."

The single gotcha: it needs H3's audio VAE wired on H3 Encode. The video VAE handles frames; audio references go through a separate VAE, and if it's not connected the node still passes the audio along but nothing conditions on it. That's the classic "I wired audio and nothing happened" failure, and it's the first thing to check.

When to reach for it

Use H3 Audio for a bare soundtrack - a voice to mimic, a music bed, an effect you want referenced - when there's no video clip attached to it. If the audio belongs to a video reference (a clip and its own sound), put it on H3 Video instead, which keeps the pair bound together; the pack exists because ComfyUI's core slot-numbering lets clip and soundtrack drift apart. If you have several audio references, H3 Inputs' audios slots hold them all on one node, and they're interchangeable with this one on the wire.

Install

Part of the NynxzH3 pack - ComfyUI Manager → "Nynxz H3", or:

cd ComfyUI/custom_nodes
git clone https://github.com/Nynxz/ComfyUI-NynxzH3

Restart. Pure Python, no dependencies, ComfyUI 0.30.0+ required. The audio VAE is a separate download from the video VAE - make sure it's in your vae folder before you expect audio references to do anything.

The one-liner

Wire your audio in, say <Audio 1> in the prompt, and above all connect the audio VAE on H3 Encode. That's the whole job - and honestly, for such a small node it does it well.