Music for video games: writing a loop that survives four hundred plays

Game music has a requirement no other medium has: it must survive being heard four hundred times. A trailer plays once. A menu loop plays every time someone opens the game, and if there is a clever hook in bar three, that hook becomes torture by hour two.

That single constraint should change how you brief every track. Most generated game music fails not because it sounds bad but because it was written to impress on the first listen, which is exactly the wrong target.

Here is how to brief music for each part of a game, what the licensing actually requires, and where AI generation genuinely fits into an indie budget.

The five music jobs in a game, and how they differ

Menu and title. Sets tone before anything has happened. Heard constantly, often while the player is doing something else like reading settings. Wants atmosphere and a slow loop, not a melody that demands attention.

Exploration and ambient. The workhorse. Long, low-information, harmonically stable. This is where "does not annoy after the fortieth loop" is the entire spec.

Combat and tension. Rhythmic, driving, and it needs to sit under sound effects rather than fight them. The single most common mistake is writing combat music with a busy midrange, which is exactly where every weapon and impact sound lives.

Victory, defeat, level complete. Short stings, five to fifteen seconds. Structure matters more than composition here.

Cutscene and narrative. The one place you can write like a film composer, because it plays once at a specific moment.

Briefing all five the same way is why a soundtrack sounds like a playlist rather than a score.

Writing the brief so a loop survives repetition

Four things go in the prompt that most people leave out.

Say it is a loop. Ask for a seamless loop with no cold open and no big ending, because a track that builds to a climax cannot loop.

Ask for low information density. Fewer events per bar, less melodic movement, more texture. A memorable melody is a liability in an exploration track.

Reserve the midrange. For anything that plays under gameplay, ask for a mix that leaves space in the mids: bass and pads carry it, and the busy detail lives high or low.

Set a mood floor, not a mood peak. Combat music that starts at maximum has nowhere to go when the boss appears.

A workable brief for an exploration bed:

Ambient orchestral underscore, sparse, contemplative, sustained string pads,
soft mallet percussion, low woodwind, no lead melody, wide reverb space,
mids kept clear for dialogue, seamless loop, 72 BPM, instrumental only

Notice how much of that is negative space: no lead melody, mids kept clear, instrumental only. Game music is often defined by what you tell it not to do, and vague adjectives cannot express that. The prompt format that makes this possible is the same six-slot structure that works everywhere.

For the combat counterpart, keep the instrumentation family identical and change only the rhythm and register. That is how a soundtrack sounds like one world instead of five.

Structure control matters more here than anywhere

A generated track that fades out in the wrong place is unusable in a game, because you will loop it.

Two levers. First, the section tags in the Lyrics field set the arrangement explicitly, which is how you control length and structure rather than hoping. Second, generating instrumentals is its own technique rather than just leaving the lyrics box empty, and a stray vocal in a menu loop will ruin a build.

For stings, generate longer and cut. Asking for eight seconds gives the model no room to shape anything; asking for thirty and trimming to the best eight gives you a usable sting.

Licensing: the part that ends projects

This is where indie developers get hurt, and it is entirely avoidable.

Commercial rights are not automatic. On Suno, they attach to songs made while you hold an active paid plan, never retroactively. A track made on the free tier cannot ship in a commercial game, and subscribing after you built the whole soundtrack does not fix it. The full rule is worth reading before you generate a single asset.

Copyright and licence are different questions. A platform can grant you the right to ship a track while the track itself may not be copyrightable, because under the US Copyright Office's position prompting alone is not authorship. For a game that matters if you ever need to stop someone else using your soundtrack. The distinction is one most developers have never had explained to them.

Storefronts and platforms have their own rules. Console certification, storefront policies and publisher agreements may require disclosure of AI-generated assets. Check before submission, not after.

Streamers will play your game. If your soundtrack triggers content matching on Twitch or YouTube, streamers get muted VODs and your game gets a reputation for being unstreamable. Original generated music avoids this by not being in the matching databases, which is a genuine advantage over library music that thousands of others also use.

What this costs versus the alternatives

Rough shape of an indie game's music budget:

Commissioned composer. Best result, real ownership, and it prices out most solo developers for a full soundtrack.

Stock and library music. Cheap, cleared, and every other game in your genre used the same tracks. It also carries the streamer-matching risk above.

Generated music. Cheapest per track, unlimited variations, and it requires you to do the creative direction the composer would have done. That is the trade: you save money and spend judgment.

The honest recommendation is a hybrid. Generate the ambient and exploration beds, where volume matters and nobody is listening closely, and commission or hand-craft the two or three moments that carry the game emotionally.

A workflow for a full soundtrack

1. Write the world's sound palette first. Three to five instruments that appear across every track. This is what makes a score cohere.
2. Build one exploration bed and live with it. Loop it for an hour while you work. If it survives that, it will survive your players.
3. Derive the others from it. Same palette, changed rhythm and register for combat, thinned for menus, resolved for victory.
4. Generate in batches per mood, not one track at a time. Consistency is easier to hear when you compare four takes side by side.
5. Export stems where you can. Layered mixing, where the combat layer fades in over the exploration bed, is the technique that makes game audio feel responsive, and it needs separated parts.
6. Keep every prompt. A year later when you need three more tracks for the DLC, the prompt is the only thing that gets you the same world back.

That last point is the one people learn painfully. Store the string, not just the audio.

Where to start if you have no music vocabulary

Most developers can hear what they want and cannot name it. That gap is the entire problem, and it is why the prompt comes out as epic fantasy music and the result sounds like a stock trailer.

Our catalogue exists to close it: 1,960 style recipes written as instruments, era, production and tempo, filterable by family and mood, so you find a sound and adjust it rather than inventing vocabulary you do not have. Ten are open free. The rest is $5 a month or $19 once.

FAQ

Can I use AI-generated music in a commercial game? Under Suno's terms, yes, if the track was made while you held an active paid plan. Free-tier tracks cannot be monetised and the licence is not retroactive.

Will it get my game's streamers muted? Original generated music is not in content-matching databases, so it avoids the problem that popular library tracks create.

How long should a game loop be? Long enough to delay the sense of repetition, usually 90 seconds to three minutes for exploration, and short stings of five to fifteen seconds for events.

Can I get seamless loops? Ask for a loop with no cold open or ending, then trim at a bar line. Some manual editing is normal.

Do I need to disclose AI-generated audio? Storefronts and publishers increasingly ask. Check the specific policies for your platforms before submission.

What about adaptive or layered music? You need stems. Plan for stem export from the start rather than trying to separate a finished mix later.