I’ve been thinking about a simple question that turns out to be surprisingly hard to answer: what does a composer actually deliver?
For most of the history of game audio, the answer was straightforward. Files. A finite set of them — an exploration theme, a combat theme, a stealth cue, maybe some stingers for key story moments. The composer’s job ended at the handoff. What happened after that, in-engine, was someone else’s problem.
That model is cracking. Not because AI has suddenly become good enough to replace human creativity — it hasn’t, and I’ll come back to that — but because the underlying concept of a finished audio asset is starting to look like the wrong unit of measurement for interactive media.
A score that actually listens. A soundtrack that’s different every time not because of a random shuffle, but because it’s responding to the player…
A Score That Listens Back
Here’s the shift in plain terms. In the old model, a game’s music is essentially a pre-recorded artifact. The engine cross-fades between states, and if the transitions are clunky — if you can hear the loop reset — that’s a production problem with no clean fix. You’ve shipped the files. That’s what there is.
In the emerging model, the composer doesn’t deliver finished music. They design a musical system — a set of rules, parameters, harmonic relationships, emotional boundaries — and an AI acts as a real-time performer within that system, responding to what’s actually happening on screen: where the player is, how tense the situation has become, what narrative beat just landed. The music composes itself, moment to moment, from the palette the composer defined.
That’s not a small change. It inverts the creative relationship between composer and output. You’re no longer authoring a piece of music. You’re authoring the conditions under which music can exist. Whether that feels like liberation or loss probably depends on what you love most about composition in the first place.
Key Takeaways
- Game composers are shifting from delivering finished audio files to designing musical systems — sets of rules and parameters that AI performs within, in real time
- Platforms like Mubert, AIVA, Wondera, and Plus Music are taking fundamentally different approaches: utility generation, creative collaboration, catalogue + AI hybrid, and adaptive licensing respectively
- The “50% cost reduction” headline is real in narrow use cases but hides new costs: specialised talent, software licensing, and engine performance overhead
- AI handles mood competently — it struggles with nuance; the gap between functional and moving is still a human space
- Copyright remains genuinely unsettled: in the US, fully AI-generated content can’t be copyrighted, and contract clauses protecting composer work from AI training are now standard practice
- The hybrid model — AI for procedural variation, humans for core creative vision — is where most studios are landing, whether they call it that or not
The Players Building This Space
A few platforms are actively working in this area, each making a different bet on what matters most.
Mubert functions as an API-first utility — generating continuous, non-repetitive ambient loops in real time, designed for environments that need to breathe without ever hitting a hard reset. Think vast open-world games where atmosphere is everything. Music as infrastructure, rather than music as expression.
AIVA takes a different angle. It positions itself as a creative collaborator — generating emotionally structured, compositionally coherent pieces in orchestral and cinematic styles, integrating into production pipelines through middleware like Fmod and Wwise. Their argument is that the output should feel composed, not generated. Whether it does is a matter of ongoing debate.
Then there’s Wondera, playing the long game with a platform strategy: AI generation combined with a claimed library of millions of pre-cleared tracks, aimed at studios that need both creative flexibility and legal certainty. And Plus Music, which doesn’t generate anything — instead, it builds Unity and Unreal plugins that make existing human-composed music adaptive. Their product is essentially: we’ll make your composer’s work respond to the game, without any of the generative legal risk. That’s not a trivial offer right now.
What the Cost Argument Misses
There’s a version of this story being told in boardrooms that goes: generative audio cuts production costs by over 50%. That number is plausible in narrow circumstances — ambient procedural content in a 100-hour open-world game, for instance, where the alternative is commissioning hours of non-repetitive orchestral music from human talent. For that specific problem, the math probably works.
But the headline figure quietly ignores a few things. Real-time neural audio generation puts computational load on the game engine — a performance consideration that doesn’t disappear because the marketing materials don’t mention it. Bridging the gap between AI engineering and game audio design requires specialised talent that’s currently rare and expensive. And the platforms themselves aren’t free; you’re trading variable composer fees for fixed software licensing and R&D costs. For larger studios with technical infrastructure, that trade might make sense. For smaller teams, the calculation is murkier than the pitch suggests.
What’s more interesting to me than the cost argument is the creative one. Because the cost conversation treats this as a replacement story — AI instead of composers — when the more honest framing is that it’s an expansion of what composition can mean.
The Human-Made Premium
Not everyone is buying the vision, and the pushback deserves more than a footnote.
The core artistic critique is this: AI-generated music is competent at mood. It can do tense. It can do sad. It struggles with nuance — with the specific emotional texture of something like wary optimism, or the particular bittersweet quality of a theme that needs to carry ten hours of narrative weight. That gap between technically functional and genuinely moving is where composers live, and no generative system closes it reliably right now.
There’s a market signal worth noting here. An upcoming RPG, Chronicles Medieval, publicly reversed its decision to use generative AI for its soundtrack after community feedback. The studio now describes all its artistic assets as “100% handcrafted.” That’s a branding decision as much as a creative one — a bet that human-made reads as premium to their audience. Whether that holds as the technology matures is an open question, but it tells you something about where listener expectations are right now.
The legal landscape adds another layer of friction. Around 63% of game developers report serious concern about copyright ownership of AI-generated content — reasonably so. If an AI generates a theme, the question of who owns it has no clean answer yet. The developer who wrote the prompt? The platform that ran the model? The artists whose work trained it? In the US, fully AI-generated content currently can’t be copyrighted, since copyright requires human authorship. Contracts are now being written with specific clauses to prevent composer work from being used as training data. The industry is improvising its legal framework in real time, and that uncertainty alone is enough to make many studios hesitant.
Composing the Conditions
So where does this leave composers, practically?
The hybrid model — AI handling procedural variation, ambient generation, and seamless transitions; humans focusing on core themes, emotional architecture, and creative vision — is the path most studios are actually walking right now, whether or not they label it that way. It’s less dramatic than either the “AI replaces composers” headline or the “nothing will change” counter-narrative. It’s a genuine expansion of the toolkit, with real creative implications that are still being worked out.
What it asks of composers is a shift in thinking. Not just: what should this music sound like? But: what are the rules of this musical world? What can happen here, and what can’t? How does tension build when an AI is the one deciding note by note? That’s closer to how a film composer thinks about a score’s internal logic — or how a generative instrument asks you to design a system rather than play a part — than how most game composers have traditionally worked.
The best version of this isn’t AI as a cost-cutting measure. It’s AI as a way to make game music finally match the interactivity that games have always promised. A score that actually listens. A soundtrack that’s different every time not because of a random shuffle, but because it’s responding to the player.
Whether the tools are there yet is another question. But the concept is right. And when the concept is right, the tools tend to catch up.
