AudioX Prompt

The least impressive node in this pack is also the most honest one. AudioX Prompt is a single multiline text box that hands a string to the next node. No API calls, no hidden processing, no key. If you've used the CLIPTextEncode-style prompt inputs in ComfyUI before, you already know how this works - this is just the version wired up for the AudioX conditioning chain.

So why does it exist? Because the model this pack wraps, AudioX (HKUST's ICLR 2026 "anything-to-audio" diffusion transformer), takes a text prompt alongside its video and audio conditions, and the pack's Condition node expects the prompt to arrive as an actual STRING output, not as a bare widget. If you tried to drag a wire from a random text node you'd fight type mismatches all the way down. AudioX Prompt exists so the chain has one clean, predictable source for the prompt. Fair enough.

Inputs and outputs

One input, one output:

• text_prompt (STRING) - the text description. The default is "Generate music for the video", which is a decent hint at the intended use case, but you'll want to write something that actually describes the audio you want: "gentle rain on a tin roof, distant thunder" or "orchestral swell, hero walking toward camera".
• text_prompt (STRING, out) - the same string, passed through untouched.

Wire the output into the text_prompt input of Condition. That's the whole job. The value is also directly readable in the node, so it doubles as documentation of what you asked for, which is genuinely useful when you're A/B-ing prompts on the same video.

A few honest caveats

The default prompt is a music prompt, and music generation is where AudioX is strongest - the paper's best results are in text-to-audio and text-to-music. If you want foley-style effects for a video, write the prompt like a sound designer, not a song title. And note that because the pack's Condition node forces all conditions in (video, text, audio), this prompt box is never optional - the chain won't build without it. It's also a plain ComfyUI STRING, so you can wire it from a wildcard or prompt-randomizer node instead of typing by hand. Nothing about this node stops you.

Installing AudioX

Install the pack via ComfyUI Manager (search "ComfyUI-AudioX") or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/Yuan-ManX/ComfyUI-AudioX.git
cd ComfyUI-AudioX
pip install -r requirements.txt
conda install -c conda-forge ffmpeg libsndfile

Then restart ComfyUI. The dependency list is heavy - the pack vendors AudioX's whole stack (pinned pandas==2.0.2, descript-audio-codec, laion-clap, wandb), so if you run ComfyUI in a managed venv or one-click portable, the pip install step can fight your environment. The model itself downloads automatically from Hugging Face the first time LoadAudioXModel runs; the README's manual wget steps are optional.

The bottom line

That's it. Install the pack once, and this node is the part that never breaks, because it does almost nothing. Treat it as the friendly face of a pack whose other nodes are considerably more opinionated.