Most people meet AIVA the same way. They type “cinematic sad piano” into the prompt box, hit generate, and get back something that sounds like a film score being played through a wall. Then they close the tab and go back to paying for stock music.
The problem isn’t the engine. AIVA is a composition tool wearing the clothes of a one-click generator, and once you treat it that way the results change completely. Here’s the exact workflow I use to build a 60-second track for a video, end to end, in under an hour.
Before you open the editor, write your brief in one sentence
Nail down three things: the mood, the instrumentation, and the length you actually need. Not “sad” but “restrained, single piano, no drums.” Not “epic” but “orchestral strings with a slow build and a peak at 45 seconds.”
This matters because the length field in AIVA is not a suggestion. If you ask for three minutes and then cut the first 60 seconds out, you will almost certainly lose the intro and keep the tail. If you know the track is going under a 60-second explainer, generate 60 seconds.
For my running example, the brief is a two-minute documentary segment about urban beekeeping. Warm and slightly curious, acoustic guitar and soft strings, with a one-minute intro sting at the top.
Step 1: Start from a preset, then break it
AIVA ships with hundreds of style presets grouped by mood and genre. The temptation is to use them as-is. Don’t. Use one as a starting point and change at least two things about it.
- Pick the preset closest to your brief. For the beekeeping piece I’d start somewhere in the acoustic or film category.
- Change the tempo. Presets tend to sit at comfortable mid-tempos. Dropping 110 BPM down to 92 instantly makes the same preset feel less generic.
- Change the key. C major is where most defaults live. Moving to A minor or D major gives you a different emotional colour for free.
- Trim the instrumentation. If the preset includes a drum kit and you don’t want drums, remove them rather than hoping the mix hides them.
Ten minutes of this gets you something that no longer sounds like a demo. If you want a broader sense of what the tool can actually do before committing to a workflow, this overview of how AIVA is transforming music creation is a useful orientation.
Step 2: Write a prompt that describes structure, not vibes
AIVA’s text-to-music model responds to concrete musical language far better than to adjectives. “Happy song” gets you mush. “Fingerpicked acoustic guitar, gentle, 90 BPM, building to full strings in the second half” gets you something you can actually place under a scene.
The four-part prompt formula
Instrument, texture, tempo feel, shape. For the beekeeping track that reads:
“Warm fingerpicked acoustic guitar, light reverb, unhurried, gradually joined by soft legato strings, ending on a resolved chord.”
That last clause matters more than it looks. Tracks that end on a resolved chord are dramatically easier to fade out than tracks that stop mid-phrase.
When to skip the prompt entirely
If you already have a melody in your head, upload it. AIVA accepts MIDI files and will harmonise, orchestrate, and extend your idea while keeping the melody intact. This is the single biggest quality jump available inside the tool, because the memorable part stays yours and the machine handles the filler. Even four bars of melody is enough to work with.
Step 3: Edit the sheet music, not the audio
This is the step most people skip, and it’s the reason their tracks still sound machine-made. AIVA gives you a notation view alongside the audio, and a few small edits go a long way:
- Delete a note that clashes with the melody line
- Drag a chord one beat later to create breathing room
- Change a single instrument’s dynamics so it sits further back in the mix
- Copy a four-bar phrase and paste it after itself to extend a section
You don’t need to read music fluently. You need to spot the one bar that sounds off and remove a note from it. That’s a thirty-second fix that saves you from regenerating the whole piece.
Step 4: Lock the duration and the loop point
Set the target length before you export. If you’re scoring a video, generate to the exact length of the sequence. A 58-second clip under a 60-second scene will loop awkwardly, while a 61-second track gives you a clean handle to fade.
If you need a bed for something longer, generate 30 seconds and set the loop points so bar one and bar 31 line up. The editor lets you mark section boundaries, which is far more reliable than cropping in your video editor afterwards and hoping nobody notices the seam.
Step 5: Export deliberately, especially on the free plan
The free tier caps downloads, so the order you export in matters. Get the structure right, export once in MP3 for a rough placement, and only spend the higher-quality export once you’ve watched the cut through twice and the track has survived it.
If you’re working inside that cap, this guide to turning AIVA’s free tier into real soundtracks breaks down how to stretch a limited number of downloads into a finished piece. The habit it teaches, exporting a reference mix first and committing later, is worth keeping long after you upgrade.
Step 6: Place it and cut to the music
Drop the file into your timeline, then nudge your cuts so they land on beats or phrase endings rather than on arbitrary frames. Two small habits make an AI track feel handmade:
- Cut on the downbeat of a new section instead of halfway through a phrase
- Ride the music under dialogue by 6 to 10 dB rather than muting it entirely
If the track still isn’t sitting right, sometimes the honest answer is that it’s the wrong tool for that particular job. AIVA is strongest on instrumental beds and orchestral cues. For tight, loopable background music under fast edits, a dedicated video-scoring tool like Beatoven.ai is often the faster route. Matching the tool to the cue saves more time than forcing one to do everything.
Common mistakes that make AIVA output sound cheap
Generating at the wrong length
Three minutes for a 45-second slot means you’re using a quarter of the track, and probably the wrong quarter. Generate to fit the slot.
Leaving the default tempo
Most presets sit at a safe mid-tempo. Nudging it even 10 BPM in either direction is the cheapest way to make a track feel like it was written for your edit instead of for everyone’s.
Never opening the notation editor
The audio-only workflow is faster, but the notation view is where you fix the one bad bar. Skipping it is why so many AIVA tracks get abandoned at the 80% mark.
Exporting before you’ve placed it
Get a rough MP3, watch the cut, then commit. You’ll waste far fewer downloads and far less time.
What a finished pipeline actually looks like
Start to finish, the beekeeping sting goes like this: brief written in one sentence, preset chosen and retuned, prompt describing instrument and shape, one melodic idea uploaded as MIDI, two bars edited in notation, exported at 60 seconds to the frame, and placed with cuts landing on phrase endings. That’s about forty minutes of real work, and the result fits the scene because you built it to fit the scene rather than hunting for something close.
AIVA isn’t going to write a hit for you in one click, and it isn’t trying to. Direct it like a session musician: give it a brief, a structure, and a fixed runtime, and it stops behaving like a slot machine. Push it into a bigger automated pipeline and tools like Dify for building AI apps without heavy code can handle the orchestration around your audio once the music workflow itself is settled.