How to Generate Music with AI: A Practical Workflow
A seven-stage workflow for generating music with AI, from writing the brief to checking the export, with exact settings and a full worked session.
By MusicVivo Editorial TeamUpdated 13 min read
Quick answer: To generate music with AI, write a short brief that says where the track will be heard, pick a genre, set tempo, energy, vocals and length, and render. Then change one thing at a time until it fits, generate 2 to 4 variations, and check the export in your editor before you publish. The whole loop takes 15 to 30 minutes for a first project.
This walkthrough is for anyone who has a concrete job in hand, such as a video that needs a bed, a game that needs a menu theme or a friend who deserves a song, and wants a dependable process instead of random clicking. It assumes no music training.
You will leave with a seven-stage routine, exact starting values for each control in the MusicVivo AI music generator, and one full session written out so you can see how the edits accumulate.
Key takeaways
- 01Start every project with a brief that names the use, not just the style.
- 02Set the big switches first: genre and vocals. Fine-tune tempo and energy after the first render.
- 03Edit one phrase or one slider per render so you learn what each change does.
- 04Use variations only once a prompt is close; they choose between good options, they do not fix a bad brief.
- 05Always test the downloaded file in its real context before calling it finished.
The seven-stage workflow at a glance
Here is the full loop. The rest of the article explains each stage with concrete numbers.
Write the brief
One or two sentences: what the music is for, how it should feel, and anything it must avoid.Pick the big switches
Choose a genre preset and decide vocals: none, soft or lead.Set tempo, energy and length
Use a BPM number if timing matters; otherwise stay near the genre default.Render a first draft
Listen once all the way through without stopping.Make one focused edit
Change a single phrase or slider and render again.Generate variations
When a prompt is close, ask for 2 to 4 versions and choose.Export and check
Download, drop it into your project and test on two kinds of speakers.
Stage 1: write a brief, not a wish
A brief answers three questions: where will it be heard, what should it feel like, and what must it not do. Style words come second.
Compare two briefs for the same job, a podcast intro.
Wish: "cool upbeat intro music"
Brief: "10-second podcast intro for a weekly tech news show, confident and bright, punchy synth bass and tight electronic drums, ends on a clean hit so the host can start talking, no vocals"
The second version tells the model the length, the function, the ending and the absence of vocals. It also names two instruments rather than leaving everything open. That is the pattern to copy: function, feeling, two or three instruments, one constraint.
If you get stuck finding the right descriptive words, keep the text-to-music prompting guide open in another tab. It groups vocabulary by genre, texture and energy.
Stage 2: the big switches, genre and vocals
Two settings change the result more than any others.
Genre sets a cluster of defaults at once: drum style, instrument palette, typical tempo and mix character. MusicVivo starts from eight core styles (Lo-fi, Synthwave, Pop, Rap, Cinematic, Jazz, Rock and Kids), and you refine inside each with text. Pick the closest preset even if your target is a subgenre, then name the subgenre in the prompt: "Cinematic" plus "minimal piano and low strings, documentary feel."
Vocals decide whether you are making a song or a bed. Set them to none for anything that sits under speech. Choose soft for a gentle sung texture, and lead for a full song where the voice is the focus. With vocals on you also choose male or female and the language, from more than 20 options. If you plan to paste lyrics, our guide on formatting lyrics for AI shows how to lay out section tags so the structure comes through.
Stage 3: tempo, energy and length
Now set the dials that shape pacing. Starting values matter because they anchor the first render.
- Meditation or sleep bed55–70
- Study or lo-fi background70–90
- Vlog or travel opener90–110
- Product promo or ad110–128
- Workout or sports edit128–150
- Drum and bass energy160–175
Tempo. If your edit is cut to a beat, or the music has to match a dance routine, type the number in the prompt as well as setting the slider, for example "102 BPM." Otherwise stay inside the range for your use. We explain why exact numbers help later editing in setting BPM and key.
Energy. Think of energy as density: how many parts are playing and how hard. For anything under speech, start in the lower third. For a hype edit, start high. Mid values give a gentle build.
Length. Ask for a little more than you need. A 45-second scene is easier to fill from a 60-second render, because you can trim a clean ending. Tracks can run up to four minutes. On the Free plan tracks are 30 seconds, and paid plans unlock full length.
Stage 4 and 5: the first render and the focused edit
Render, then listen once from start to finish without scrubbing. Write down your reaction in plain words: "too busy in the middle," "drums too loud," "ending is abrupt." Each note becomes one edit.
The rule for this stage is simple: one change per render. Here is a translation table from common complaints to the single edit that usually fixes them.
| What you hear | One edit to try |
|---|---|
| Too busy under speech | Lower energy one notch, or add "sparse arrangement" |
| Melody competes with the voice | Add "no lead melody, pads and rhythm only" |
| Feels sluggish | Raise tempo by 6 to 10 BPM |
| Wrong instrument dominates | Replace it by name: "nylon guitar instead of piano" |
| Ending stops dead | Add "gentle fade on the final chord" or "ends on a sustained chord" |
| Vocals swallow words | Shorten lines to 6 to 9 syllables and render again |
Keep a short log. Even three lines in a notes app ("v1 too busy, v2 lower energy better, v3 added fade, keep") makes the next project faster.
A worked session from start to finish
Job: music for a 90-second YouTube video showing a weekend camping trip. Light narration in the first 30 seconds, then a montage with no speech.
v1 brief: "acoustic instrumental for a camping trip video, warm and adventurous, acoustic guitar, hand percussion, 100 BPM, no vocals." Genre preset: Pop. Energy: middle. Length: 100 seconds.
Reaction: nice feel, but the opening is as busy as the montage, so it fights the narration.
v2, one change: add "starts quiet and sparse for the first 30 seconds, then builds into a fuller groove." Everything else the same.
Reaction: the build works. The montage section now feels a bit thin.
v3, one change: raise energy one notch.
Reaction: the build is right and the montage lifts. The last chord is cut off.
v4, one change: add "ends on a held, ringing guitar chord."
Reaction: this is the one. Now ask for variations to choose the best take.
| Sparse intro (narration) | 16 bars |
| Build | 8 bars |
| Full groove (montage) | 16 bars |
| Lift | 8 bars |
| Outro chord | 4 bars |
At 100 BPM in 4/4 time, one bar lasts 2.4 seconds, so 16 bars of sparse intro covers about 38 seconds, enough room for 30 seconds of narration plus a breath. That is the kind of arithmetic that makes tempo numbers worth typing.
Bars to seconds
At any tempo, one bar of 4/4 lasts 240 divided by the BPM, in seconds. At 120 BPM that is 2 seconds; at 80 BPM it is 3 seconds.240 / BPM
seconds per 4/4 bar
2.4 s
one bar at 100 BPM
38 s
16 bars at 100 BPM
Stage 6: variations, used at the right moment
Variations are several renders of the same prompt. They differ in melody, drum fills and small arrangement choices while keeping the style. They are most useful once a prompt is already close, because then every option is a reasonable candidate.
The number you can request per prompt depends on plan: 1 on Free, 2 on Starter, 4 on Creator and Pro. Each variation uses credits for its length, at one credit per second of audio. A 60-second track rendered as four variations uses 240 credits, so it pays to reach a good prompt with single renders first.
- 1 render60 credits
- 2 variations120 credits
- 4 variations240 credits
When choosing between variations, listen to all of them at the same volume, then pick on the hardest moment of your project: the point where speech starts, or where the montage peaks. The nicest-sounding version in isolation is not always the one that fits.
Stage 7: export and check in context
Download the files and put them where they will live. With MusicVivo you get an MP3 and an .lrc lyrics file timed line by line plus a 1024 by 1024 cover, on every plan. Free renders are 128 kbps, while paid plans deliver original quality. Our guide to AI music exports covers how to use each file.
Before you call it finished
- Play it under the real voiceover or game footage, not on its own.
- Listen on headphones and on a phone speaker.
- Check the first and last two seconds for clicks or abrupt cuts.
- Confirm your plan covers the use: personal on Free, commercial on Starter and above.
- Save the final prompt and settings next to the project file.
If the track is going into a monetized or client project, read can you use AI music commercially for how licenses usually work, and keep a record of which plan you were on when you rendered. Our article on keeping license records suggests a simple format.
Try it
Run the seven stages on your own brief
Write one sentence about where the music goes, set vocals and tempo, and hear a draft in about a minute.
Five workflow mistakes and how to avoid them
Starting with variations. Requesting four versions of a weak prompt gives you four weak tracks and burns four times the credits. Reach a prompt you like with single renders, then branch.
Rewriting the whole prompt after each render. If v2 shares almost nothing with v1, you lose the thread of what improved. Keep a stable core sentence and edit only the tail. For example, keep "warm acoustic instrumental for a camping video, acoustic guitar and hand percussion" fixed and experiment only with the phrase about structure or ending.
Ignoring the slider when the text disagrees. A prompt that says "slow and dreamy" while the tempo slider sits near the top asks for two things. When results feel confused, check that every control points the same way as the words.
Judging a bed on its own. A background track that sounds dull on its own is often perfect under a voice, and a track that sounds exciting solo can bury the narration. Always audition beds under the real dialogue, even a rough phone recording of it.
Forgetting what worked. The prompt that finally landed is the most valuable thing you made in the session, more than the audio file, because it can produce the next ten tracks. Save it with the settings: genre, tempo, energy, vocals, length.
The same workflow for a song with lyrics
The seven stages hold when vocals are on, with two additions.
First, decide the lyrics source in stage 1. Either describe a theme ("a lighthearted song about a dog who steals socks, playful, for a family video") or paste your own words. When you paste, mark sections clearly:
[Verse]
Every morning there's a sock gone missing
Underneath the couch I hear him hiding
[Chorus]
Biscuit, Biscuit, bring it back
One more sock in your secret stack
Second, add a vocal check in stage 7. Read along with the render and make sure every word lands. If a line gets rushed, it is usually too long for the melody; trim it by two or three syllables and render again rather than changing the music settings. Lines of 6 to 10 syllables sing most cleanly at mid tempos around 90 to 110 BPM.
Reusing the workflow across projects
Once a prompt works, save it as a template with blanks: "[length] instrumental for [use], [mood], [instrument 1] and [instrument 2], [BPM] BPM, no vocals." Next time you fill the blanks instead of starting from zero. Creators who publish weekly often keep five or six templates, one per recurring segment. The AI music workflow for content creators shows how to batch these for a publishing schedule.
If you are moving on to songs with lyrics rather than beds, the companion guide on how to make a song with AI covers lyric writing, structure tags and vocal choices in the same step-by-step style.
People also ask
How do I generate music with AI step by step? Write a brief that says where the music will be heard, choose genre and vocals, set tempo, energy and length, then render. Change one thing per render until it fits, then generate variations and check the export in your project.
How long does it take to generate a song with AI? On MusicVivo a two-minute track usually renders in about 60 to 90 seconds. The full workflow, with a few comparisons and checks, takes 15 to 30 minutes for a beginner.
What settings matter most when generating AI music? Genre and vocals change the result most, then tempo and energy. Length matters for fitting a scene but rarely changes the character of the track.
Should I write my own lyrics or let the AI write them? Both work. Paste your own lyrics with section tags when you have specific words or names, or describe a theme and let the generator write them, then edit and render again.
Related reading
- What Is an AI Music Generator? A Beginner's Guide
- Text-to-Music Prompting: Genre, Mood and Tempo Words
- Setting BPM and Key When Generating AI Music
- An AI Music Workflow for Content Creators
- How to Make a Song with AI
- Background Music for Videos
Product and company names mentioned in this article are trademarks of their respective owners. MusicVivo is an independent product and is not affiliated with, sponsored or endorsed by any of them. Comparisons are based on publicly available information at the time of writing; check each provider's own site for current features, pricing and terms.