H3 Continuum Load Audio

MiniMax H3 generates audio natively - dialogue, sound effects, music, baked into the video latent rather than stitched on after. That's the whole reason the "H3 Continuum Load Audio" node exists in this pack, and the reason you have to think about which audio socket you're feeding. The node itself is boring-good: it's ComfyUI Core's Load Audio with a bypass-friendly Enable Audio switch and an H3 category sticker. One audio file in, one AUDIO out. The interesting part is what you connect that output to, because Continuum treats audio references and audio drivers completely differently.

Driving Audio vs. Reference Audio

On the H3 Continuum Sampler V3.8 there are two audio paths, and they are not interchangeable:

• Driving Audio - you want this when the supplied track should be the finished video's sound. It guides generation as native H3 conditioning and is selected unchanged as the final output, bypassing generated audio and seam processing.
• Reference Audio - this only conditions generation (voice timbre, style, a sound you want the model to riff on). Generated audio remains the final audio; the reference never replaces it.

Same loader feeds both - the socket you wire it into decides the contract. For a source video with its own sound, the standard move is to route the video's frames to Video Guide Frames and its audio to Driving Audio so the clip's original track survives to the end.

What it is under the hood

H3EasyLoadAudio subclasses ComfyUI Core's LoadAudio. The Enable Audio toggle is a frontend extension that flips the node into ComfyUI's native bypass - OFF behaves like an unconnected optional input, so you can disable the audio path in a reusable workflow without deleting the node or its cable. There's no audio processing happening here; this is a loader, full stop.

Inputs and outputs

One required audio file combo, one AUDIO output. That's the whole surface. Keep the file where ComfyUI can find it (your input folder) and pick it from the dropdown.

Installing

Installing this node is just installing its pack:

cd ComfyUI/custom_nodes
git clone https://github.com/ukr8b3g-cmyk/ComfyUI-H3-Continuum.git

Restart ComfyUI, or search "H3 Continuum" in ComfyUI Manager. No pip packages to install - requirements.txt is deliberately empty; the pack uses the PyTorch and safetensors that ship with ComfyUI (0.32.0+).

Gotchas

Don't hook one audio source into both Driving Audio and Reference Audio expecting the same result - driving replaces your generated audio, reference doesn't, and the pack's docs warn against mixing the bundled reference path with the legacy single Reference Audio (Optional) socket on the sampler. And the standing model-level caveat applies here too: the MiniMax H3 weights this feeds come from HuggingFace under a community license that excludes the US, EU, UK and South Korea, and no audio or model files ship with the pack - that 45-second track with the exact room tone you want is yours to supply.