What instruments should I name in an Indian music prompt?
Name the instruments that actually carry the sound, since the model falls back on generic genre defaults if you don't. Common choices: tabla (hand drums, sharp slap plus deep tone), sitar (plucked string with a buzzing resonance), dholak or dhol (hand or stick drums for rhythm), harmonium (reedy hand-pumped keyboard), tanpura (a drone instrument that hums in the background) and tumbi (a short-necked twangy string used in bhangra).
- Tabla and dholak for rhythm in film songs and devotional pieces
- Dhol and tumbi for bhangra and dance tracks
- Sitar or harmonium for the main melody or chord bed
- Tanpura only if you want a drone under a slow, meditative piece
What tempo and key work for this style?
Match tempo to mood: 60 to 90 bpm for emotional or romantic film songs, 90 to 110 bpm for slow builds and devotional pieces, and 128 to 140 bpm for bhangra or item-song energy. Minor keys suit drama and longing; major or Lydian (a bright, slightly floating major scale) colours suit celebration and wonder. State the bpm and key as separate words rather than folding them into a mood description.
How should the vocal be described?
Describe the voice by timbre and gender rather than naming a singer: warm female vocal, confident male vocal, breathy female vocal, gentle devotional male vocal. Reliable vocal words include male/female, whisper, breathy, raspy and belting; words like "smooth" or an exact age are less predictable. If you want Hindi, Punjabi or Turkish lyrics sung, note that Lyro Music routes sung lyrics, including Turkish, to Google's Lyria 3 Pro, while vocal chops, tempo-locked beats and audio-input work go through ACE-Step 1.5.
How do I keep the prompt from sounding generic?
Use 4 to 7 concrete descriptors covering rhythm, melody instrument, harmony instrument, voice and mood, and give one tempo. Avoid piling on more than three mood words, since ACE-Step 1.5 and similar models degrade when moods contradict each other, for example pairing "romantic" with "aggressive". A tight, specific prompt outperforms a long descriptive paragraph.
| Weak | Stronger |
|---|---|
| "Indian song, happy" | "Bollywood-style film song, tabla rhythm, sitar melody, warm female vocal, joyful mood, 100 bpm, major key" |
| "Punjabi music" | "Punjabi bhangra, dhol drum, tumbi line, brass stabs, male vocal, 132 bpm, major key" |
What do models get wrong with this style, and how do I fix it?
Models often flatten tabla and dholak rhythms into generic pop drums, blur sitar and harmonium into a vague "world music" pad, and default to English-sounding vocal phrasing even when you ask for Hindi or Punjabi lyrics. Fix this by naming each instrument separately rather than saying "Indian instruments", stating the language explicitly next to the vocal description, and keeping the mood list short so the model doesn't average out the drama.
Lyrics skeleton
Original, for structure
[Verse] Under the same old sky tonight I'm counting stars, waiting for your light [Chorus] Come back to me, let the music play Every step of yours lights up my way [Verse] Roads we walked, still they call my name Heart still beats the same old flame [Outro] Hold me close, let the night ignite
Common mistakes
- Saying "Indian instrumental" with no named instruments, which invites generic world-music defaults: name tabla, sitar, dholak or harmonium directly.
- Naming a specific singer or film to get "that sound": describe the voice (warm female vocal, confident male vocal) and mood instead.
- Mixing bhangra energy words with romantic-ballad tempo: pick one direction, since 130 bpm plus "soft romantic" produces a muddled result.
- Leaving out tempo and key: without them the model guesses, and film-song and bhangra styles sit at very different bpm ranges.
- Writing lyrics directly in the prompt/style box: put lyrics in the lyrics box and keep the style box to sound description only.