AudioX Prompt Helper

The hardest part of text-to-audio is often the text. Sound diffusion models want a specific kind of description - named sounds, materials, context - and getting that right by hand is fiddly. AudioX Prompt Helper is the pack's built-in prompt-builder: you type a rough idea, it returns a polished audio prompt (plus a negative prompt and a summary), ready to feed into any generation node.

It's not a generation node itself. It produces strings, which is exactly why it's useful - you can build and inspect your prompt before committing to a 250-step diffusion run. Wire its enhanced_prompt output into any AudioX generator's text_prompt input and iterate on the building blocks instead of the raw sentence.

How it works

Give it a base_prompt and it layers on improvements in order:

• template - 16 presets across music_*, nature_*, urban_*, and action_* categories (nature_forest, urban_cafe, action_impact, …). Appends the preset's scene description.
• style_modifier (optional) - cinematic, realistic, ambient, dramatic, peaceful, energetic. Prepends the matching style terms.
• enhance_automatically (default on) - the same enhance_audio_prompt keyword-tagging used across the pack ("rain on a window" → "natural environmental rain on a window sound").
• add_quality_terms (default on) - tacks on "high quality, clear, well-defined".

Inputs & outputs

• base_prompt - your rough idea.
• template / enhance_automatically / add_quality_terms / style_modifier / negative_prompt (optional) - the building blocks above. The negative_prompt defaults to "muffled, distorted, low quality, noise" and is passed through unchanged.
• enhanced_prompt - STRING, wire into a generator's text_prompt.
• negative_prompt - STRING, the (mostly unused, for now) negative.
• prompt_info - STRING summarizing which modifications were applied.

Install & gotchas

Standard pack install, no model weights needed:

cd ComfyUI/custom_nodes
git clone https://github.com/lum3on/ComfyUI-StableAudioX.git
cd ComfyUI-StableAudioX
pip install -r requirements.txt

Honest notes:

• The quality terms are filler. "high quality, clear, well-defined" is generic padding the model mostly ignores. It's harmless, but if prompts feel bloated, turn add_quality_terms off.
• The negative output has no consumer yet. AudioX's enhanced nodes log negative prompts but don't apply them (the tooltips say implementation pending), so this output is for future use or other models.
• Templates assume a scene. nature_ocean is opinionated - if your base prompt already says "harbor," the appended template may fight it. Prefer none when you're confident in the base.

For sketching ambience prompts or standardizing how your team writes audio prompts, it's a genuinely handy utility - it turns "water sounds" into something the model actually responds to. Just remember it's a prompt factory, not a silver bullet.