Instrumental prompts for ACE-Step: keeping the vocals out

ACE-Step is a music model trained primarily on tracks with vocals, so its default assumption is that vocals belong somewhere. Getting a clean instrumental requires a few specific moves.

Z.Tools5 min de lectura
Og Image

ACE-Step v1.5 can return humming, syllables or a buried vocal line when you ask for an instrumental. Getting a clean instrumental takes a few specific settings, and none of them are obvious from the UI. This article did not generate audio: it relies on the AI Audio to Audio tool's documentation and the ACE-Step musician's guide, and the prompt habits below are untested suggestions.

The two dials that actually keep vocals out

Two settings need to line up.

The first is the lyrics field. The tool's documentation says ACE-Step generates an instrumental when the lyrics field is empty or contains only structural tags. A tags-only field may still produce a vocalise (not tested here), so a fully empty field, with no tags and no whitespace, is the safer choice.

The second is the vocal language picker. It lists 50 languages plus an option labeled "Instrumental / Auto", which the documentation recommends for cleaner instrumental results. How the model uses that option internally was not verified.

The documentation recommends both settings together; how reliably they keep vocals out was not measured.

Prompt language that signals instrumental

The prompt can help or hurt. Three habits to try:

Name the instruments, not the singers. A prompt that lists piano, drums, bass and synth pad and never mentions a vocal element gives the model less reason to add one. Nouns such as "the song" or "the track" may imply a vocal.

Include the literal phrase "no vocals." The tool's documentation uses it at the end of an instrumental example prompt. Its effect on top of an empty lyrics field was not measured.

Avoid genre prompts that imply vocals. A "pop", "ballad" or "folk song" prompt may make the model expect vocals, since those genres usually have them. "Underscoring", "BGM", "instrumental score" and "video game music" name settings where instrumental versions are common.

Two dials and the prompt anatomy for clean instrumentals

Three example prompts

None of these was generated for this article, so none is verified to give a clean instrumental. Each is meant for an empty lyrics field and the Instrumental / Auto language.

Prompt one: lo-fi study music

Lo-fi hip-hop instrumental, dusty boom-bap drums at 88 BPM, jazzy electric piano comping in seventh chords, mellow upright bass walking lines, vinyl crackle, late-night focus mood, no vocals.

Why it is built this way: the genre is named in a way that fits instrumentals, two harmonic instruments (electric piano, upright bass) and drums are named, vinyl crackle adds texture, the mood is described without reference to a song or a singer, and "no vocals" closes it.

Prompt two: synthwave underscoring for a video

Mid-80s synthwave underscoring, 110 BPM, retro analog arpeggio, gated reverb snare, sidechained pad, four-on-the-floor kick, cinematic synth melody on top, instrumental score, no vocals.

Why it is built this way: "underscoring" frames it as music for video, where instrumentals are common, and "cinematic synth melody on top" names the lead element, so no vocal lead is needed.

Prompt three: orchestral video game scoring

Orchestral video game scoring, slow strings rising, soft horns, timpani swells on the climax, harp arpeggios in the verse, no vocals, no choir.

Why it is built this way: "video game scoring" carries instrumental connotations, and the instruments are named and spread across sections. "No vocals, no choir" excludes wordless choral pads, a plausible failure for orchestral prompts that was not tested here.

Patterns to avoid

Three patterns that may work against an instrumental request (none tested here).

Saying "instrumental" once in the middle of a long prompt. Among 200 characters of vocal-implying language it may count as one feature among many. Put it near the front or near the end.

Leaving lyrics filled with bracket tags only. [Verse] and [Chorus] with no lyrics may be read as cues to add vocalise or hummed lines that follow the section structure. The tool's documentation allows tags-only lyrics for instrumentals, but an empty field is the safer choice.

Setting a specific vocal language while requesting an instrumental. "English" together with an instrumental request gives contradictory signals, and a quiet vocal layer is a possible result.

Generador de Covers con IA

Generador de Covers con IA

Convierte una canción en un cover con IA con otro género, voz y arreglo, conservando la melodía y la letra.

A note on MiniMax Music Cover

MiniMax Music Cover is the wrong tool for this job. MiniMax describes the model as preserving the source melody, so an instrumental request from a vocal source may come back with hummed or sung-syllable lines that follow the original singer's contour (not tested here). Use a stem splitter, or ACE-Step from scratch.

Checklist

Before you generate an instrumental on ACE-Step:

The lyrics field is empty, with no text, no tags and no whitespace. The vocal language is set to "Instrumental / Auto." The prompt names instruments rather than song elements. The phrase "no vocals" appears in the prompt at least once. The genre or purpose framing (underscoring, BGM, scoring, video game music) implies an instrumental context.

None of this guarantees a vocal-free result. Generate a few seeds on Turbo (a four-minute track costs $0.024) and listen to each.

Generador de Covers con IA

Generador de Covers con IA

Convierte una canción en un cover con IA con otro género, voz y arreglo, conservando la melodía y la letra.

  • ai
  • music
  • ace-step
  • instrumental

Sigue leyendo