AceStepPromptGen

Writing a good ACE-Step prompt is nothing like writing a good image prompt. The model responds to dense, technical descriptions - BPM, specific instruments, production texture - and "lo-fi chill beats" gets you a sad little sine wave. This node is the cheat sheet: a dropdown of 228 professionally-written style templates, each one a full sentence of the exact kind of detail ACE-Step's text encoder wants.

What it actually does

Pick a style and it hands you a ready-made prompt string. For example, the "Afrobeats" template reads something like "Afrobeats at 105 BPM with layered log drums, shekere, talking drum, Afro-fusion guitar licks, deep 808 sub, and syncopated hi-hats." That's the whole trick - a curated library of style-specific prompt text, written by someone who clearly tuned it against the model, so you start from a prompt that's already at the quality level most people iterate toward after twenty renders.

The dropdown is enormous and genuinely global: 60+ electronic subgenres (Acid Techno through Breakcore), deep Brazilian coverage (Samba, Bossa Nova, Forró, Funk Carioca), rock/metal, jazz, classical, and world music. This is the pack's underrated asset.

The inputs that matter

• style - the 228-choice dropdown. This is 95% of the node.
• extra - free text appended after the style template. Want "with a female vocalist" or a mood tweak? Type it here. Empty is fine; the template alone is a complete prompt.
• voice_style (optional) - 18 pre-written vocal hints, split into female (natural, breathy, powerful, ethereal, soulful, deep), male (plus tenor/baritone), blended, and a robotic_vocal option. These append phrases like "natural female voice with micro pitch variation, soft breath, realistic vibrato" - directly aimed at the AI-vocal problem this pack obsesses over.

Outputs

Two strings. prompt is the final assembled text (template + extra + voice hint) - this is what you feed into your ACE-Step text conditioning. template is just the bare style text, useful if you want to see the base prompt or reuse it elsewhere with your own modifications.

Install

Manager (search "JK AceStep Nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/jeankassio/JK-AceStep-Nodes.git

Restart ComfyUI. No extra downloads - the styles are baked into the code. It pairs naturally with the pack's KSamplers and lyrics nodes, but the prompt string it emits works with any ACE-Step workflow, including ComfyUI's stock text-to-music nodes.

Where people trip up

The "extra" text is concatenated, not blended - there's no weighting or re-ordering, so don't try to fight the template with it. And voice_style defaults to none (auto), which is honestly the right call: voice hints are a fix for vocal artifacts, not a flavor you want every render. Use them when a vocal sounds wrong, then drop them once you've found a prompt that holds. For 90% of sessions, this node is: pick a style, run.