QWE AI Academy Hot take: most AI music tutorials get prompting backwards. They tell you to lead with a genre – “write a lo-fi track” – and then layer mood on top. But genre is a container, not a feeling. “Lo-fi” can sound nostalgic, melancholic, sleepy, or focused depending on what you put inside it. If you want a specific mood, you should describe the mood first and let the genre serve it, not the other way around.
This guide walks through how to actually prompt for AI music with specific genres and moods, which tool gives you the most control for the price, and the licensing fine print that catches people three months in. No “AI is changing the music industry” preamble – straight to the tactics.
Why genre-first prompting underdelivers on mood
Look at how Google structures its own prompt guidance. Lyria’s official prompt template splits a prompt into five separate fields: Genre/Era, Tempo/Rhythm, Instruments, Vocals, and Lyrics. Genre and Era leads with a specific style or musical era; Tempo and Rhythm sets energy; Instruments asks for specific sounds or solos; Vocals specifies gender, timbre, and range; Lyrics describes topic or provides custom text with structure tags. Notice that “mood” isn’t its own slot – it emerges from the combination.
That’s the trick. Mood is a downstream result of tempo + instrumentation + vocal texture. If you write “sad lo-fi” you’re letting the model guess. If you write “slow tempo, dry intimate acoustic guitar, breathy female vocal in low register, sparse beat” you’re engineering it.
The 4-layer prompt that actually controls mood
Here’s the structure I use across Suno, Lyria, and ElevenLabs. It works because every modern text-to-music model parses these dimensions independently before fusing them.
- Mood anchor – one or two emotional adjectives (melancholic, triumphant, anxious, warm)
- Tempo + rhythm feel – BPM range or descriptor (“driving 140 BPM,” “slow swaying ballad”)
- Instrumentation – 2-3 specific instruments, including their texture (“distorted bassline,” “dry acoustic guitar”)
- Genre + era as the container – last, not first (“in the style of late-90s trip-hop”)
Example prompt for Suno or Lyria:
Anxious, claustrophobic mood.
Mid-tempo around 95 BPM with a stuttering hi-hat.
Features a detuned Rhodes piano, sub bass, and
breathy whispered female vocal in the low register.
Late-90s trip-hop style.Compare that to “sad trip-hop song” and you’ll hear the difference on the first generation. The four-layer prompt gives the model fewer degrees of freedom to wander.
Walkthrough: building a specific mood in Suno
Suno is the most accessible starting point. As of April 2026, the free tier gives you 50 credits per day, Pro is $10/month for 2,500 credits, and Premier is $30/month for 10,000 credits with commercial rights. Here’s the actual flow.
- Open suno.com and switch to Custom Mode. The default “simple” mode hides the style and lyrics fields where you actually need control.
- In the Style field, paste the 4-layer prompt above. Don’t put it in the lyrics box – that’s a common mistake.
- For lyrics, use structure tags like
[Verse],[Chorus],[Bridge], and[Outro]. Structure tags genuinely change how the model paces the song – without them you often get a one-section loop. - Generate two variants. Suno produces two by default per credit batch. Listen to both before regenerating – sometimes the “worse” one has the mood you wanted but a weaker hook you can fix in v2.
- Use Extend on the better take to add length while keeping the mood locked in. Re-prompting from scratch usually drifts.
Pro tip: If a mood keeps slipping toward “upbeat” no matter what you write, the culprit is usually the major key default. Add an explicit key request like “in D minor” or “minor key throughout” to the style prompt. Models will obey it more often than not.
The licensing and credit traps nobody warns you about
This is where most tutorials wave their hands and say “AI music is royalty-free!” It’s more complicated.
Suno’s UTC credit reset. The free plan gives you 50 credits per day, those credits do not roll over, and the reset is aligned to UTC – not your local midnight. If you live in California and start a session at 10pm local, your credits already reset hours ago. Plan generation sessions around UTC if you’re squeezing the free tier.
No retroactive commercial rights. Commercial use is only granted on paid plans; free-tier songs are personal use only. Paid subscribers receive commercial licensing rights for songs created during their subscription. That song you made on the free tier last month doesn’t become commercial just because you upgraded today – you’d need to regenerate it under your paid subscription.
The Warner deal is changing downloads in 2026. In late 2025, Suno and Warner Music Group announced a settlement and partnership. The arrangement is meant to clear a path for licensed AI models in 2026, along with tighter controls on how audio leaves the platform – free-tier users on the new licensed system will be limited to playback and sharing, not full file downloads. If you’ve been relying on free MP3 exports, that workflow may not survive the year.
The lawsuit context. Worth knowing: Sony Music Entertainment, Universal Music Group’s UMG Recordings, and Warner Records filed federal copyright infringement lawsuits against Suno and other tools via the RIAA, alleging use of copyrighted music to train the models. The Warner side has been resolved via partnership; Sony and Universal status varies as of this writing. Use AI music for commercial work with eyes open.
Tool comparison: which one for which mood
Different tools have different sweet spots. Here’s an honest comparison based on current public specs (verified April 2026 – pricing changes often, confirm before subscribing):
| Tool | Free tier | Max length | Best for | Commercial use |
|---|---|---|---|---|
| Suno | 50 credits/day | Extends with re-prompt | Vocal pop, rap, full-song demos | Paid plans only |
| Lyria 3 Pro (via Artlist) | Trial credits | 30 sec – 3 min | Cinematic, prompt-precision work | Per Artlist license |
| ElevenLabs Music | Trial credits | Varies by plan | Multilingual vocals, ad jingles | Self-Serve allows commercial except film/TV/large studio games |
| SOUNDRAW | Limited preview | Editor-driven | 30+ genres, stem export | Worldwide perpetual license while subscribed |
| Canva AI Music | Plan-dependent | 180 seconds | Background score for Canva designs | In-design only – no raw audio export |
Sources: Artlist’s Lyria page, ElevenLabs Music, Canva’s AI Music FAQ, and SOUNDRAW’s pricing page. Watch for the Canva detail specifically: you’re allowed to include AI-generated music in commercial digital content as part of your own designs and exports, but you can’t download the raw audio file and use it separately.
For precise mood control, Lyria’s separated prompt fields give you more deterministic results. For quick vocal-driven songs, Suno wins on speed and quality. For background scores where you want to stay well clear of the ongoing lawsuit cloud, SOUNDRAW positions itself as an alternative – though verify their current licensing terms before committing.
Common pitfalls (and the fixes that actually work)
The mood drifts every regeneration. Lock it with explicit constraints – key, BPM, and at least one named instrument. Mood adjectives alone aren’t enough signal.
Vocals sound generic. Specify range and texture: “airy female soprano,” “raspy male baritone with vibrato.” Generic prompts get generic vocals.
The song structure is mush. Use bracket tags in lyrics. [Verse 1], [Pre-chorus], [Chorus], [Bridge], [Outro]. Without them, models often skip the bridge or repeat the chorus four times.
You hate every output. The prompt is probably too long. Models start ignoring tokens after a certain length. Cut to ~60-80 words max in the style field.
FAQ
Can I use AI-generated music on YouTube monetization?
Depends on the tool and plan. Suno’s paid plans grant commercial rights for songs made while subscribed; free-tier songs are personal use only. Always check the current terms before monetizing.
Why does the same prompt give wildly different results each time?
Because text-to-music models sample from a probability distribution – there’s randomness baked into every generation. Say you prompt “melancholic piano ballad” three times: you might get one waltz-like piece, one slow rock ballad, and one ambient drone. The fix is over-specifying constraints (key, BPM, instruments by name) so the distribution narrows. You’re trading creative surprise for predictability – pick which one you need that day.
Is AI music actually copyright-safe to release commercially?
It’s evolving. Some tools train on licensed or in-house catalogs, while others have faced major-label lawsuits with partial settlements. The safest path right now: use a paid commercial plan from a tool with clear licensing terms, keep records of your generation date, and avoid prompts that name living artists. Don’t assume “royalty-free” on a marketing page means “litigation-proof.”
Next step: pick one mood you want to nail – say, “hopeful but tense” – and write three versions of the 4-layer prompt for it. Generate one of each in Suno’s free tier. Compare. The differences between your three prompts will teach you more about mood control in 20 minutes than any tutorial.