|
•••
|
Most creators who try AI music for the first time prompt it the way they prompt AI image tools. They write prose. They use adjectives. They describe the feeling they want and hope the model produces something that matches. The output is usually disappointing. Then they conclude AI music is not ready yet, when the real problem is that the prompt format does not fit how music models read input.
AI music tools reward a different prompting style entirely. Where image models want descriptive prose, music models want structured metadata. Where image prompts use adjectives, music prompts use numeric constraints. Where image generation is forgiving of vague style references, music generation collapses without a specific genre, tempo, and production register. The skill transfers in spirit. The format does not.
I have been testing Suno, Udio, and Stable Audio for the last several months, mostly for podcast intros, short video soundtracks, and the kind of background music that a small brand asks for casually. Here is what actually works.
|
|
The Template
Seven specific ingredients.
|
|
The structure that consistently produces strong output has seven slots. Genre, tempo, mood, instrument, vocal style, era, and reference. Each one does a specific job. Together they give the model a precise enough target to produce a coherent track rather than a meandering interpretation.
[GENRE] + [TEMPO] + [MOOD] + [INSTRUMENT] + [VOCAL STYLE] + [ERA] + [REFERENCE]
A working example looks like this:
Indie folk, 92 BPM, melancholic, fingerstyle acoustic guitar, whispered female vocals, 2010s indie style, in the spirit of early Bon Iver.
Twenty-six words. Seven specific ingredients. No wasted adjectives. Suno or Udio takes that prompt and produces something that sounds intentional rather than generated. The same prompt with "sad song with guitar" produces something that sounds like AI music. The difference is not the model. It is the precision.
|
|
Why Each Slot Matters
What the seven ingredients actually do.
|
|
Genre anchors the entire structure. Chord progressions, rhythm patterns, and production aesthetic all flow from this single word. Be specific. "Lo-fi hip hop" and "boom-bap hip hop" produce different tracks.
Tempo in BPM is more useful than words like "slow" or "fast." Suno responds to numbers. 78 BPM, 92 BPM, 120 BPM each carry specific feeling. Vague tempo language produces inconsistent results across generations.
Mood guides melody and harmony decisions. Skip basic emotions. Use "melancholic," "nostalgic," "haunting," "euphoric," "bittersweet," "triumphant" instead of "sad" or "happy."
Instrument tells the model which signature sound carries the track. One specific instrument gives the track identity. "Fingerstyle acoustic guitar" is stronger than just "guitar."
Vocal style shapes the performance. "Whispered female vocals," "raspy male tenor," "choir harmonies in falsetto" are all instructions the model can interpret. Vocal gender alone is too thin.
Era sets the production register. "2010s indie style" produces different mixing decisions than "1970s analog warmth" or "early 2000s alternative rock." The era word does a lot of work in one slot.
Reference is the optional final ingredient. "In the spirit of" plus an artist or aesthetic gives the model a target without trying to clone anyone specifically. Naming actual artists triggers copyright restrictions, so use stylistic descriptors instead.
|
|
Bracket Tags
The underused structural instruction.
|
|
The single thing nobody tells beginners is that the brackets in Suno and Udio prompts are not decorative. They are real instructions. The model reads [Intro], [Verse], [Chorus], [Bridge], [Outro] as structural commands and shapes the track around them. Same goes for production tags like [Vinyl crackle], [Warm Rhodes piano], [Brushed snare], [Reverb-heavy guitar].
A full prompt using bracket tags looks like this:
[Intro] Vinyl crackle, warm Rhodes piano, lazy 70 BPM, ambient pad
[Verse] Soft male vocal, fingerpicked acoustic, brushed snare
[Chorus] Layered female harmonies enter, gentle string swell
[Bridge] Strip back to piano and vocal, intimate
[Outro] Vinyl crackle returns, fade to silence
This is closer to a producer's session notes than to a creative writing prompt. The track Suno generates from a structured tag prompt is dramatically more usable than the same idea expressed as prose. If you are not using bracket tags yet, this is the highest-leverage change you can make to your music prompting workflow.
|
|
Three Working Prompts
Copy, swap the theme, generate.
|
|
PROMPT 01 LO-FI BACKGROUND TRACK
Lo-fi hip hop, 70 BPM, melancholic, warm Rhodes piano with vinyl crackle, no vocals, 2010s lo-fi era, in the spirit of late-night focus playlists. [Intro] Tape hiss, soft Rhodes chord progression. [Loop] Lazy boom-bap drums enter, subtle bass groove, no melodic lead. [Outro] Drums drop out, Rhodes returns alone, fade.
Use for podcast intros, YouTube background music, focus playlists, study content.
|
|
PROMPT 02 INDIE FOLK SONG
Indie folk, 92 BPM, nostalgic, fingerstyle acoustic guitar, whispered female vocals, 2010s indie style, in the spirit of intimate bedroom recordings. [Intro] Solo guitar fingerpicking. [Verse] Soft vocals enter, sparse arrangement. [Chorus] Brushed drums and bass join, vocal harmonies layer in. [Bridge] Back to solo guitar and vocal. [Outro] Full arrangement returns briefly, then fades to single guitar note.
Use for emotional short films, brand storytelling videos, About-page atmosphere.
|
|
PROMPT 03 CINEMATIC TRAILER
Epic orchestral, 120 BPM, triumphant, layered strings with deep brass, no vocals, contemporary trailer score era, in the spirit of modern blockbuster cues. [Intro] Single piano motif. [Build] Strings enter quietly, low percussion taps. [Drop] Full orchestra hits, brass swells, taiko drums. [Resolve] Strings sustain, piano returns, slow fade.
Use for brand launch videos, product reveal trailers, dramatic social media moments.
|
|
Three Practical Notes
Things I learned the hard way.
|
|
Exclusion prompts work. If your output keeps including elements you do not want, add them as negatives in the Style field. "No rap, no electronic drums, no distorted guitar" is a legitimate instruction. The model honors exclusion language consistently, and it is one of the least-used features in the platform.
Genre mashups are stronger than expected. "Celtic folk meets hip-hop" or "jazz with trap drums" or "ambient electronic with country vocals" all produce coherent tracks. The Suno v5.5 model handles over 1,200 genres, so niche combinations land more reliably than they have any right to.
Iterate in batches before judging. Suno produces two versions per generation, so two creations produce four interpretations of the same prompt. Compare them before deciding the prompt does not work. The first version is rarely the keeper, exactly the same dynamic as AI image work. Generate a small set, then choose.
|
|
Which Tool for What
Suno, Udio, Stable Audio, Lyria.
|
|
Suno v5.5 is the most reliable all-rounder for songs with vocals. Arrangements feel like songs rather than loops, hooks tend to land, and lyrics come out clean. Pro tier at ten dollars a month gives full licensing rights for YouTube and client work.
Udio is the closest competitor and generally has slightly stronger audio fidelity in hip hop, R&B, and modern pop. Long-form generation is more coherent. The interface is simpler, the feature set is narrower. Worth a side-by-side test if vocal tracks are your primary use case.
Stable Audio is a different animal. It is built for instrumental music, sound design, and short loops rather than full songs. The right pick for sound effects, background pads, and short cues for video work where you do not need a complete arrangement.
Google Lyria 3 Pro is a no-cost option that works inside the Gemini app. Quick generation, three-minute cap, no licensing rights on the entry tier. Useful for prototyping a track idea before committing to Suno or Udio for the final version.
|
|
The honest summary is that AI music has matured faster than most creators realize. The tools are good enough to ship right now for podcast intros, video soundtracks, brand cues, and the kind of background music that small clients ask for casually. The skill that separates working creator-musicians from people who gave up on AI music after one bad output is just understanding that the format is different. Numeric constraints. Bracket tags. Genre layered with reference. Production language over creative writing.
If you have been writing music prompts the way you write image prompts, try the template once. The difference in output is the difference between an idea sketch and a track you would actually use.
|
|
•••
I am putting together a music prompting pack with twenty tested templates across lo-fi, indie folk, cinematic, electronic, hip-hop, jazz, and ambient. Each one structured with bracket tags. Each one ready to paste.
Want it when it ships? Reply with send me the music pack and I will get it to you.
|
|
A QUESTION FOR YOU
What kind of track would you generate first?
Reply and tell me what you are making. I will send back a structured prompt tuned for your specific use case.
If this resonated, forward it to a creator who has been frustrated with AI music.
|
|
Until next time,
Luxe Prompting
|
|
Luxe Prompting
AI Image Generation for Creators
|
|
