Load Dia Audio

Load Dia Audio is how you tell the Dia TTS model whose voice you want. Dia isn't a generic text-to-speech bot with a pick-a-voice dropdown; it's a dialogue model that takes a reference clip of a voice and matches it. This node is the front door for that reference - it reads a WAV file off disk and hands it to Dia TTS as an audio_prompt, which then drives the voice, tone, and emotional read of whatever you generate.

It does exactly one job, and does it without drama. The file gets read with soundfile, converted to mono if it's stereo, normalized to float32, and passed through as a ComfyUI AUDIO value - the standard (sample rate, waveform) tuple every audio node in this pack speaks.

Inputs and output

One input, one output:

• file_path - a plain text field, default ./example_audio_prompt.wav. There's no file browser here; you type a path. And heads up: that default file doesn't ship with the pack - it's a placeholder, so you'll get a "file not found" error until you point it at something real. The path is relative to wherever your ComfyUI server process runs (usually the ComfyUI root, not your workflow folder), so when in doubt, use an absolute path.
• audio (output) - wire it into Dia TTS's audio_prompt input.

That's the whole node. It doesn't transcribe, doesn't analyze, doesn't cut. Feed it a good clip, get good voices.

Installing

It's part of ComfyUI-Dia, so it installs with the pack - ComfyUI Manager (search "ComfyUI-Dia") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Yuan-ManX/ComfyUI-Dia.git
cd ComfyUI-Dia
pip install -r requirements.txt

Then restart ComfyUI.

Common issues

• "Audio file not found" - almost always a path problem: wrong spelling, wrong directory, or a relative path that resolves against the ComfyUI root rather than where you think it is. Absolute paths end this class of bug.
• It reads the wrong audio - make sure you're pointing at the reference voice, not your output or some random sound effect. The clip sets voice and mood; a noisy or multi-speaker clip gives you mush.
• One reference, one gender - a quirk people hit with Dia generally: a single reference clip locks you into voices of the same gender as the reference. If your scene needs a male and a female character, you'll need a separate reference for each. That's a model limitation, not a node bug.
• Keep it a WAV - soundfile can technically read other formats, but WAV is the tested path here. Convert first and save yourself the headache.