Which AI soundtrack generator can score a film or a game you are making?

Here is the music list for a twelve-minute short film, the kind an editor writes on a sticky note the night before picture lock:

1M1  Opening titles      0:00-0:48   sparse piano, cold, no melody yet
1M2  Kitchen argument    2:10-3:05   low strings, pulse that tightens
1M3  The drive           4:40-6:20   needs to swell exactly at 5:52 (headlights)
1M4  Silence             6:20-8:00   no music
1M5  End credits         10:30-12:00 a song, with words, she sings it

And here is the same list for a small game:

Menu loop          plays for as long as the player sits in settings
Exploration bed    loops for 20+ minutes, must not wear out
Combat layer       must start on the beat over the exploration bed
Victory sting      4 seconds
Credits song       vocal, 3 minutes

No single AI soundtrack generator delivers either list from top to bottom. Some tools are good at the long, quiet bed. Some give you stems you can switch in and out. One or two will put a voice on the credits. Almost none will hit a headlight at 5:52 for you, because that is editing, not generation. The useful question is not "which tool is best" but "which line of my list does this tool cover, and what do I do about the lines it does not".

Loops, cues and songs are three different orders

Before comparing tools, sort your list into three piles, because every tool is quietly built for one of them.

Loops are game music in its purest form. They have no beginning and no ending in the musical sense, they repeat, and they succeed when nobody notices them on the fortieth pass. A track that builds to a climax is useless as a loop, however good it sounds the first time.

Cues are film music. They start and stop at specific moments, they follow a scene, and they often have to land something (a cut, a door, a look) on a precise second. A cue is judged against the picture, never on its own.

Songs have words and a voice. In a film that is usually the end credits or a scene where a character hears a radio. In a game it is the credits, a trailer, a tavern bard, an in-world station.

If you already know you are working in Suno and need to brief a menu loop or a combat track so it survives four hundred plays, our guide to music for video games covers the prompt side in detail. This page is for the step before: you are still choosing the tool, or you suspect you need two of them.

What each tool actually hands you

Everything below was checked against each company's own pages at the time of writing. Plans and limits change, so treat the numbers as a snapshot and re-check the pricing page before you pay.

AIVA is the closest thing on this list to a composer you can argue with. It generates across more than 250 styles, lets you edit what it generated, accepts an audio or MIDI influence to steer it, and exports MIDI on every plan. MIDI is the point: you can pull the notes into your own software, change the orchestration, cut a bar, shift a swell to land on your headlight. The catch sits in the licence. On the free and Standard plans AIVA's pricing page states "Copyright owned by AIVA"; only the Pro plan reads "Copyright owned by YOU" with full monetization. Free also means no monetization at all and a required credit. If the film is going to a festival or the game is going on sale, that line decides your plan, not the feature list. AIVA's own pages do not say it sings, and if you need words, our piece on what to use when AIVA's instrumentals are not enough picks up there.

Soundraw is built around a mixer. You choose genres, it builds a track, and then you set the length, tweak the intensity, toggle instruments, and the page says it "rebuilds your track on the spot". Its FAQ mentions editing at the bar level (mute, solo, intensity). For a game this is attractive: you can strip a track down to a quiet version and keep a full one. Stems are only on the Artist Pro plan and above; the cheaper plans are MP3 only. The pricing page says the Creator plan covers background music for apps and games. It does not name film, so if you are scoring a film, read the licence page linked from pricing before you build on it.

Beatoven.ai generates background music from a text description with its own model, plus a separate tool for sound effects, and downloads as MP3 or WAV. The licence terms deserve a slow read: the FAQ describes a "non-exclusive perpetual licence", Beatoven remains the owner of the tracks, and direct distribution to streaming platforms is not allowed. That is fine for a bed under a YouTube video. It is a question to answer before you plan a soundtrack album on Spotify.

Stable Audio 2.5 from Stability AI generates tracks up to three minutes and supports audio-to-audio and inpainting: you upload your own audio, pick a starting point, and the model continues from that context. The company says it was trained on a fully licensed dataset and pitches it at enterprise sound production, with the opening credits of a game named as one example. Inpainting is the interesting part for scoring, because you can keep a passage you like and regenerate around it.

Eleven Music from ElevenLabs writes both songs and instrumentals, lets you select a section and regenerate just that part, and exports stems on paid plans (two or four on the Creator tiers, up to six on Pro). Section regeneration is close to what a scene edit needs. But read the licence table before anything else: the self-serve plans say "All online and offline commercial use permitted, except film, TV, radio, and Studio Games". Their terms define an indie game by revenue thresholds and do not count it as a studio game, and the plan that removes the exclusion is the enterprise one. For a short film headed anywhere commercial, that exclusion is the whole answer.

Mubert Render builds tracks from a typed prompt or an uploaded image, with genres, moods, tempos and activities to browse, and has a beta option to search by a YouTube reference. The duration control is behind an upgrade. The licence is on its own page: the free tier is marked non-commercial, the paid tiers list commercial use, and the sync licence names videos, podcasts, apps and games but not film. The same page forbids distributing the tracks to streaming services and registering them in Content ID, which rules out a soundtrack album.

Suno is the one most people try first, and it covers more of the list than its reputation for pop songs suggests. There is an Instrumental toggle, so the quiet bed and the combat layer are in reach. The paid plans add stem separation (two types on Pro, three on Premier), and Premier includes Suno Studio. It is also the obvious tool for the credits song, because it sings. Commercial use rights come with the Pro and Premier plans. Suno's current terms tie them to songs you download while subscribed, within the plan's download allowance, and the help centre says subscribing later does not give retroactive rights to songs made on the free plan.

The comparison, line by line

Read the last column first. It is the one that ends projects.

Stems and layers: what a game needs that a film does not

A film plays once, in order. A game plays whatever the player does, so its music has to change state without a seam. The standard way to do that is layers: one exploration bed, and on top of it percussion, a bass pulse, a brass line, each brought in or out depending on what is happening. Audio middleware such as Wwise, or your engine's own audio system, does the switching. The music has to be built so the layers fit.

Here is the trap. If you generate the calm bed and the combat version as two separate prompts, you get two pieces that share a genre and nothing else: different tempo, different key, different swing. Crossfade between them and the player hears a gear change. Two routes avoid it.

Split one track. Generate the fullest version (the combat state), then separate it into stems. Mute the drums and brass and you have the exploration bed, already in time with the combat layers because they came out of the same take. Suno's stem separation and Soundraw's stems both work this way; Eleven Music's stem export does too on its paid plans.

Work from notes. Export MIDI from AIVA, open it in your own software, and build the layers yourself from the same notes. More work, much more control, and the only route on this list where you can move a swell by a beat.

For a film the logic is simpler. You need cues that start and end at the right second, and the generator will not know your timecode. Generate longer than the scene, then cut and fade in your editor. If you want a tool that reads the video itself, that is a different job, covered in our look at video to music tools.

Writing the instrumental prompt

Most bad soundtrack generations come from prompts written the way you would describe a film to a friend:

epic cinematic music for my game boss fight, really intense, no vocals

That gives the model a mood and a list of things it should not do, and models are poor at "not". The version that works names the instruments, the register, the shape and the role, and keeps every instruction positive:

Style:   hybrid orchestral, low brass ostinato, taiko and frame drums,
         staccato strings, dark minor tonality, driving steady pulse,
         constant intensity, open midrange, instrumental
Exclude: vocals, choir, spoken word, big final hit

Notice where the "no" went. Everything you do not want moves into the Exclude field, so the Style line only describes what should be there. That split is exactly what our guide to fixing negative prompts is about. constant intensity matters for a loop, because a track that builds has nowhere to go when it repeats. open midrange asks for room where weapon and impact sounds live.

For the exploration bed, keep the instrument family and change the shape:

Style:   hybrid orchestral underscore, sustained low strings, soft taiko,
         distant brass pads, dark minor tonality, sparse, slow-moving,
         continuous loop feel, instrumental
Exclude: vocals, choir, lead melody, drum fill

Same orchestra, so the two pieces sound like one world. If you would rather not write these by hand, the prompt generator builds a Suno style prompt on every axis, keeps it positive and moves each "don't" into Exclude, and the catalogue has worked recipes for styles like film-score, cinematic-score, hybrid-orchestral, noir-film-score, chiptune and synthwave. If a stray voice still turns up, this guide to Suno instrumentals covers why it happens.

Rights for a commercial release

This is where a soundtrack built in a weekend turns into a problem a year later. Four checks, before you generate the first cue:

1. Which plan were you on when the track was made? Suno, AIVA and Eleven Music all tie rights to the plan, and Suno says plainly that upgrading later is not retroactive by default. On Suno, download the cues you keep while the subscription is active.
2. Does the licence name your medium? Eleven Music's self-serve plans exclude film and studio games in their own words. Soundraw's Creator plan names apps and games but not film. Mubert's sync licence names games, not film.
3. Who owns it? A licence to use a track is not ownership of it. AIVA keeps copyright below Pro; Beatoven keeps ownership and grants a non-exclusive licence. Non-exclusive means somebody else may hold the same rights to similar material.
4. What does your distributor, publisher or festival ask? Storefronts, publishers and festivals can have their own rules about AI-generated assets and disclosure. Ask before delivery, not after.

Copyright in AI output is a separate question from the licence and differs by country; this explainer on who owns AI-generated music separates the two. This is not legal advice. Our own position on what we generate is on the disclaimer page.

Where sunomarket fits, and where it does not

We are an independent site. We are not Suno and are not affiliated with Suno or with any company named above. We do not make audio.

For pure underscore, the honest answer is one of the tools above. If you need notes you can rearrange, that is AIVA on the plan that gives you the copyright. If you want a quick bed with a mixer, that is Soundraw. We are not in that race.

We cover two lines of the list.

The Suno style prompts. If Suno is your engine, the prompt generator builds the instrumental Style string, puts the "don'ts" in Exclude and keeps only the instruments you name, so the calm and combat versions stay in one orchestra. The first three preset categories are free without a card; the full version and the full catalogue come with a plan.

The song with words. End credits, a trailer song, an in-world radio track, a bard. The lyrics generator takes a short brief and returns a title and a full lyric with [Verse] and [Chorus] tags, ready to paste into Suno's Lyrics field. It does not sing; the voice comes from Suno or whichever tool you pick. A brief for the credits song of a small game about a lighthouse keeper might look like this:

theme      A keeper who stays at the lighthouse after the town leaves;
           the light still turns every night for ships that never come.
           Plain words, sea and salt, no fantasy names.
POV        first
language   en
structure  ballad
hook       title-line
vocal      female
emotion    longing
age        any

One generation on SM1 costs 20 credits, which is $0.20 at 100 credits to the dollar. SM2 and SM3 run an extra pass that flags weak lines and rewrites them, at 80 credits ($0.80). New accounts start with zero credits: there is no free trial of the lyric generator. Plans are on the pricing page. Our lyrics are non-exclusive and can overlap with what other users get, so treat the result as a draft to edit, not a guaranteed one-off.

Which tool fits which project

• A festival short with a few precise cues. You need control over timing. AIVA's MIDI on the Pro plan, or any generator plus careful cutting in your editor. Check that the licence names film.
• An indie game with layered states. Generate the fullest state once, split it into stems, build layers from the same take. Suno or Soundraw on a plan with stems, or AIVA if you want to work from notes.
• A YouTube series or a trailer bed. Speed matters more than control. Soundraw, Beatoven or Suno, with the licence checked for your channel.
• Any project with a sung credits song. Suno (or Eleven Music, if its licence covers your medium) for the voice; a lyric written for the scene, from us or from you, so the words are about your story and not about generic heartbreak.

If you want the wider map of composer tools beyond soundtracks, our comparison of AI composers sorts them by how much control you keep.

The one-cue test before you pay

Do not pick a tool from a table, this one included. Pick the hardest line on your list (for most games the combat layer that must sit over the exploration bed, for most films the cue that has to land on a moment) and run only that line through two candidates, on a free tier where one exists or the cheapest month where it does not. Then answer three questions about the result: can I make it start and stop where I need it, can I get a quieter version of it in the same tempo, and does the licence on the plan I would pay for name my medium? The tool that gets three yeses on your hardest cue will handle the easy ones. The tool that wins on the easy cue tells you nothing.