Vague prompts rarely fail loudly. You get a song that technically matches what you typed and still sounds like nobody in particular. The fix is nearly always the same: swap a word that describes a feeling for two or three words that describe a sound.
Below are twelve rewrites, most of them Indian-language songs, followed by the method behind them so you can rewrite your own prompts without a list to copy from.
Why the before versions disappoint
Words like happy, sad, romantic or epic describe how a listener should feel, but a hundred very different songs can create the same feeling. A sad song could be a ghazal, a piano ballad, a slow trap track or a folk lament. When you do not choose, the model chooses the most average option, and the average option is the forgettable one.
Every after version below adds the same kinds of detail: a specific genre, instruments you can hear, a type of voice, a tempo and, often, the purpose of the song.
Indian-language rewrites
Read each pair as what people usually type first, then what gets closer to a usable song.
- Before: 'sad Hindi song'. After: 'slow Hindi piano ballad, restrained male vocal, cello entering in the chorus, 68 BPM, rainy-night heartbreak'.
- Before: 'Punjabi party song'. After: 'Punjabi bhangra-pop, dhol and tumbi over a modern 808 beat, confident male vocal with group shouts, 135 BPM, made for a dance-floor reel'.
- Before: 'devotional song for Ganpati'. After: 'Marathi Ganpati aarti, male and female chorus, dholki, taal and harmonium, brisk and joyful, home pandal recording'.
- Before: 'romantic song for my wife'. After: 'gentle Hindi acoustic love song, warm male vocal, fingerpicked guitar and soft tabla, slow 76 BPM, anniversary surprise'.
- Before: 'Bhojpuri song'. After: 'Bhojpuri wedding dance number, energetic male and female duet, dholak and banjo, fast and loud, for a haldi celebration'.
- Before: 'Tamil song'. After: 'Tamil folk-pop, spirited female vocal, parai drums and nadaswaram over a modern beat, 120 BPM, festive and proud'.
- Before: 'qawwali'. After: 'Hindi-Urdu qawwali, powerful male lead with chorus answering each line, harmonium, tabla and handclaps, building from slow to fast'.
English and instrumental rewrites
The same method works outside Indian genres.
- Before: 'happy pop song'. After: 'upbeat English synth-pop, bright female vocal with layered harmonies in the chorus, punchy drums, 118 BPM, summer road trip'.
- Before: 'rock song'. After: 'gritty garage rock, distorted guitar riff, live-sounding drums, shouted gang vocals on the hook, raw mix, 140 BPM'.
- Before: 'relaxing background music'. After: 'instrumental lo-fi, warm electric piano chords, soft vinyl crackle, brushed drums, 75 BPM, late-night study'.
- Before: 'epic music for a video'. After: 'cinematic instrumental, rising string ostinato, low brass hits, big taiko-style drums, builds steadily to a full climax'.
- Before: 'love song in English'. After: 'acoustic R&B ballad, close and intimate male falsetto, warm electric piano, finger snaps, 82 BPM'.
Lyrics before and after
If you write your own lyrics, the lyrics box needs the same care. A common before looks like this: a title line, a line saying 'Mood: sad, music: piano', and twenty lines of text with no section labels. The after keeps style notes in Style Tags, drops the title, and marks the structure with [Verse], [Chorus] and [Bridge].
Autunes helps with part of this. It moves lines such as 'Mood: sad' or 'Music: dhol, shehnai' out of the lyrics and into the style, turns labels like 'Verse 1:' or 'Mukhda' into proper section tags, and sings a line marked (x2) twice. But only you can decide where the chorus is, and a clear chorus is what makes a song memorable.
A short example of a clean after: [Verse] then 'Shaam dhali, teri yaad aayi / Khidki pe baarish ne dastak lagaayi', then [Chorus] then 'Laut aa, laut aa, yeh dil hai pukaare (x2)'. The tags give the song a shape and the repeat gives the chorus its hook.
The rewrite method in five moves
Use this on any prompt that is not working.
- Circle every feeling word and give each one an instrument or a tempo. Emotional becomes 'piano and cello, slow'.
- Make the genre more specific by one level. Indian becomes Sufi, Sufi becomes Sufi rock.
- Describe the voice: who sings, how many, and how. Soft, powerful, raspy, rap, choir.
- Add the use. A first dance, a reel hook and a temple recording need different songs even with the same words.
- Remove anything that contradicts something else in the prompt.
When the after version still misses
Change one variable and generate again. If the genre is right but the voice is wrong, change only the voice words or the vocal tag. If you change everything at once, you will not know what helped and you will spend more credits finding out.
When a take is nearly right, a variation (5 credits) gives you a fresh take of the same song. On Pro and Creator you can also fix a single sung line with section editing for 3 credits instead of regenerating the whole song.
Check the language picker before every new song. It defaults to Hindi, and an English or Tamil song generated with the picker on Hindi is a frequent reason a good prompt sounds wrong.