How to Make Video Game Music (AI Loops in Your Browser)

A vibe coder who spent the weekend wiring a browser platformer in WizardGenie, an indie three weeks into a jam build with placeholder royalty-free loops, and a hobbyist sound designer who can’t justify another $25 Splice month for a project that may never ship all land on the same problem in 2026: how to make video game music that fits the level, loops cleanly, layers into intensity stems, and doesn’t cost the rent. The AI music stack has moved from “novelty” in 2023 to “the obvious default” in 2026 — the V5.5 model in Sorceress Music Gen writes full instrumental dungeon tracks, boss-fight crescendos, and adaptive ambient pads from a prompt in under 90 seconds, and the four-mode workflow (create, extend, mashup, cover) makes those tracks usable for games instead of just YouTube videos. This post walks the honest pipeline: the five pieces every video-game soundtrack needs, the prompt anatomy that produces game-ready output, the four Music Gen modes that turn a base track into a level’s worth of cues, and the SFX pass that fills in the hits, jumps, and pickups.

What “how to make video game music” actually means in 2026

How to make video game music is a different problem than how to make music. Music is a linear artifact — a song with a beginning, a middle, and an end — that a listener consumes once, in order, with full attention. Video game music is the opposite: a non-linear, context-aware, loopable score that adapts to the player’s pace, ducks under combat SFX, swells during boss reveals, and resets cleanly when the player dies and restarts. The same player will hear the same dungeon track for forty minutes if they’re stuck on a puzzle, so the track has to loop without an audible seam and stay listenable on the twentieth pass.

That difference is why the 2026 video-game-music workflow is built on four primitives that don’t exist in linear music production. Loop points. Every track needs a clean loop seam — the last beat of bar 32 has to transition naturally into the first beat of bar 1. Stems. A boss track delivered as one mixed file is one decision; the same track delivered as drum, bass, lead, and pad stems is four decisions the game engine can mix dynamically at runtime. Tempo locking. The combat track at 128 BPM has to cross-fade into the victory sting at 128 BPM or the transition feels broken. Layered intensity. The exploration version of a track strips out the drums and the lead; the combat version adds them back. Adaptive music is what separates a finished indie game from a tech demo with placeholder loops.

The five pieces every video game soundtrack needs (title, level loops, boss, ambient, SFX)

Naming the five pieces up front gives the AI a clean prompt target for each one and keeps the soundtrack from drifting into a single ten-minute prog-rock track that fits nothing. Every shipped indie game from a one-room jam game to a 40-hour metroidvania has roughly the same five categories.

Piece 1: The title theme. A 60 to 90 second hook that plays on the main menu, the studio splash, and the credits. This is the only piece the player hears more than once in full, so it carries the project’s sonic identity. Big intro, big outro, no loop seam required.

Piece 2: Level loops. One per region or biome. 2 to 4 minutes each, clean loop seam, mid-energy. The dungeon track, the forest track, the snow level. These do the most work in the soundtrack because the player spends 80 percent of the play session inside them.

Piece 3: The boss tracks. One per boss, plus a generic mini-boss track. 90 seconds to 3 minutes, high-energy, tempo-locked to the gameplay framerate (60 BPM divides evenly into 60 fps for animation sync). Loops cleanly because boss fights run long.

Piece 4: Ambient pads. Low-volume sustained drones for cutscenes, dialogue, and inventory menus. These give the audio mix a floor to sit on so the absence of music feels intentional, not broken. 60 to 120 seconds, almost no rhythm, loops invisibly.

Piece 5: One-shot SFX and stings. Pickup chimes, level-clear fanfares, boss-defeated stings, game-over chords. Short (1 to 5 seconds), non-looping, tempo-matched to whatever track is currently playing so they don’t sound dropped-in.

The math for a six-level indie game ends up at: 1 title + 6 level loops + 4 boss tracks + 3 ambient pads + ~20 SFX. At Sorceress 10 credits per Music Gen generation and 3 credits per SFX Gen batch, that’s 140 credits + 60 credits = 200 credits total, which the 100 free starter credits cover halfway. A $20 Creator credit top-up covers the whole soundtrack with credits to spare.

Step 1 — Prompt anatomy: how to write a game-ready music brief

The Music Gen prompt is where most first-time video-game-music attempts go wrong. The default reflex is to type “dungeon music” into the prompt box and hope for the best. The model can do better with five extra words and a clearer structure. The prompt anatomy that produces consistent game-ready output has six parts.

Genre. Be specific. “Chiptune” not “retro.” “Orchestral hybrid” not “epic.” “Synthwave” not “80s.” Chiptune alone covers eight subgenres; pin the one that fits.

Mood. Two adjectives, no more. “Tense and methodical” for a dungeon. “Triumphant and propulsive” for a boss. “Wistful and patient” for a hub town.

Instrumentation. Name the three or four lead instruments. “Square-wave lead, triangle bass, NES-style noise hi-hat, 4-bit drums.” The model latches onto specific instrument names better than abstract textures.

Tempo. Lock the beats per minute in the prompt. 120 BPM for standard combat, 90 BPM for exploration, 140+ BPM for chase sequences. The model honors explicit BPM tags far more reliably than vague “medium tempo.”

Structure. “Loopable A-B-A.” “No intro, no outro, designed to repeat seamlessly.” “Build to climax at 0:45, sustain through 1:30.” The model writes longer-form structure when you give it a shape.

Negative tags. What you don’t want is as important as what you do. “No vocals, no fade-out, no key change, no orchestral hits.” The Music Gen custom mode exposes a Negative Tags field for exactly this purpose — everything in that field is excluded from the generation.

A complete prompt for a dungeon track looks like: “Chiptune dungeon exploration loop, tense and methodical, square-wave lead, triangle bass, NES-style noise hi-hat, 4-bit drums, 110 BPM, loopable A-B-A, no intro no outro no fade-out.” Negative tags: vocals, key change, orchestral hits, modern drums. The custom mode also lets you set a Style Weight slider (default 0.5; push to 0.7+ to hold genre tighter) and a Weirdness slider (default 0.5; push down to 0.2 for safer output, up to 0.8 for unexpected combinations).

Step 2 — Create your first track with Music Gen (V5.5, 10 credits, 2 variations)

The Music Gen workflow at sorceress.games/music-gen opens on the Create tab, which is where every new soundtrack starts. The default model is V5.5 (verified against the MODELS array at line 376 and the default useState('V5_5') at line 444 of src/app/music-gen/page.tsx on June 11, 2026), with V5, V4.5+, V4.5, and V4 available as fallbacks. V5.5 is the right call for almost every prompt; the older versions are useful when you want a specific older sonic character (V4 has a noticeably crunchier low end that fits some retro genres better than V5.5’s cleaner master).

Toggle Custom Mode on. This unlocks the Style, Title, Vocal Gender, Negative Tags, Style Weight, and Weirdness fields. For game music, Custom Mode is non-negotiable — the default simple mode generates closer to a song with a verse-chorus structure, and the custom mode generates closer to an instrumental loop. Toggle Instrumental on for any track that doesn’t need vocals (which is most game music; vocals are for title themes and credit songs).

Paste the prompt into the Style field, paste any title text into Title, paste negative tags into Negative Tags. Press Generate. Each Music Gen run costs 10 credits (verified at MUSIC_CREDIT_COST = 10 on line 26 of src/app/music-gen/page.tsx) and produces 2 track variations (verified at the “2 variations per generation” UI string on line 2094). The two variations are not the same track twice — they are two distinct AI interpretations of the prompt, with different melodies, different drum patterns, different instrumentation choices. For game music, this is a feature: keep both, use one as the exploration version and one as the combat version of the same biome, or A/B-test them in the engine and ship the better one.

The generation runs in the background; you can fire off three or four prompts before the first one finishes. The Gallery on the right shows the tracks as they complete, with a waveform preview and a Play button. Listen on headphones, not laptop speakers — game audio mixes that sound fine on the laptop sound thin in Discord and muddy on a phone speaker.

Step 3 — Loop it with Extend, Mashup, and Cover modes

One track is the start of a level, not a level’s worth of music. The three Music Gen non-create modes (verified at the MODE_TABS array on lines 386 to 393 of src/app/music-gen/page.tsx) turn a single base track into the variable-length, transition-aware, adaptive score the level actually needs.

Extend mode. Pick any completed track in the Gallery, click Extend, pick the timestamp to continue from. Music Gen generates 30 to 60 additional seconds that pick up exactly where the base track ended, in the same key, same tempo, same instrumentation. Cost is the same 10 credits per extension, and each extension also returns 2 variations. The use case is variable-length levels: a 2-minute base loop extended twice gives 4 minutes of in-key music for a long dungeon. Extend modally to make the music breathe with the level’s pacing instead of restarting from bar 1 every two minutes.

Mashup mode. Upload two source tracks (or pick two from the Gallery), and Music Gen generates a new track that combines elements of both — the drums of one, the lead melody of another, blended into a single output. The use case is cinematic transitions: the dungeon track mashed up with the boss track becomes the corridor that connects them. The transition feels intentional because the mashup carries motifs from both ends.

Cover mode (uploadCover). Upload a reference track (yours, public-domain, anything you have rights to) and prompt Music Gen for a style cover. The model preserves the melody and structure but re-instruments to match the prompt. The use case is adaptive intensity: cover the calm version of the level theme as a high-energy combat version, and the engine cross-fades between them when combat starts. Same melody, same hooks, completely different mood — the player’s ear stitches them together as one piece of music.

The four modes together cover every game-music need: Create for the base, Extend for length, Mashup for transitions, Cover for adaptive layers. A six-level game can run on six base tracks plus 20 to 30 Extend/Mashup/Cover generations and still come in under 400 credits total.

Step 4 — Layer SFX with SFX Gen + trim in SFX Editor

Music without SFX is a soundtrack album, not a game. The hit sounds, pickup chimes, footstep loops, UI clicks, and door creaks are what makes the music feel like it’s reacting to the player. SFX Gen at /sfx-gen generates these from text prompts at 3 credits per SFX (verified at SFX_CREDIT_COST = 3 on line 24 of src/app/sfx-gen/page.tsx on June 11, 2026), with a batch mode that runs multiple prompts in parallel and a Variations control that returns 2 to 8 takes of each prompt at count * SFX_CREDIT_COST credits per batch.

The prompt anatomy for SFX is shorter than music: object + action + qualifier. “Metal sword hit on shield, sharp, short reverb tail.” “Coin pickup chime, bright, ascending major triad, 0.4 seconds.” “Heavy footstep on wet stone, soft, no echo.” The model handles physics-grounded SFX better than abstract ones, so reach for real-world references (“cardboard tearing,” “glass shattering on tile,” “dry leaves underfoot”) when the genre allows.

Once SFX Gen returns a take, the SFX Editor at /sfx-editor handles in-browser trimming, fading, normalizing, and looping. This is the step most AI-SFX workflows skip and most ship-able games require. A sword-hit SFX with a 200-millisecond pre-roll of silence sounds “late” in-game; trim the silence and it sounds tight. A pickup chime with a hard cutoff at the end of the last note clicks audibly when the game plays it; add a 50-millisecond fade-out and the click disappears. A footstep loop that doesn’t cross-fade between repeats has an audible seam; the SFX Editor’s loop point tool sets the seam at a zero-crossing where the waveform is flat, and the loop becomes invisible.

For full-track loop-seam editing, the same SFX Editor handles music too. Drop a Music Gen track in, find the cleanest bar-32-to-bar-1 transition, set the loop point at that exact sample, export. The game engine’s audio loader respects the loop point metadata, so the engine loops the track without writing custom audio code.

Step 5 — Drop the stems into your engine (engine-agnostic export)

Sorceress Music Gen exports MP3 (lossy, smaller, fine for web games) and the source audio Sound Studio handles WAV and OGG (lossless and Vorbis, the two formats every game engine reads natively). All three formats import into Phaser via this.load.audio('dungeon', 'dungeon.ogg'), into Three.js via the Web Audio API via audioLoader.load('dungeon.mp3', buffer => ...), into native HTML5 via the <audio> element, and into every desktop engine’s import pipeline. There is no Sorceress-specific format and no engine lock-in — the tracks are standard audio files.

For adaptive music, the engine-side wiring is straightforward. Load both the exploration version (Music Gen base track) and the combat version (Music Gen Cover of the same base) as separate AudioBufferSourceNode instances. Set the combat version’s gain to 0 and the exploration version’s gain to 1. When combat starts, ramp the gain crossover over 1.5 to 2 seconds — the two tracks are key-locked and tempo-locked because the Cover mode preserved both, so the cross-fade sounds like the music intensifying, not like two tracks fighting.

The full audio suite is at Sorceress Sound Studio, which bundles Music Gen, SFX Gen, Speech Gen for AI narration, and the SFX Editor as a single workspace. The credit math at $0.01 per credit means a complete six-level soundtrack with 200 SFX comes in around $5 of credits. The Lifetime tier at $49 (verified against LIFETIME_PRICE = 49 in src/app/plans/page.tsx on June 11, 2026) front-loads $50 of credits and never expires; the Creator credit pack at $20 buys 2,000 credits, which is enough for two indie games’ worth of music and SFX.

The verdict on how to make video game music in 2026

How to make video game music in 2026 is the same six-step pipeline every shipped indie soundtrack follows — prompt anatomy, base track creation, loop extension, mashup transitions, style-cover adaptive layers, SFX pass, in-engine export — with the “learn a DAW” step replaced by a browser tab. The five pieces (title, level loops, boss, ambient, SFX) cover every game from a one-room jam build to a 40-hour metroidvania. The four Music Gen modes (Create, Extend, Mashup, Cover) cover every adaptive-music need the engine throws at the score. The 200-credit budget covers a full six-level soundtrack, and the 100 free starter credits at signup cover the first dungeon’s worth at zero dollars.

The trade is honest. AI-generated music is a starting point that the developer’s ear has to finish — loop seams need trimming, key clashes between tracks need re-generating, the occasional weird drum fill needs vetoing via Negative Tags. The Music Gen output isn’t a replacement for a real composer on a triple-A budget; it’s the replacement for the placeholder royalty-free loops every indie ships with anyway. That trade favors the indie developer hard. The composer who would have spent 80 hours writing 60 minutes of original music now spends 8 hours curating and trimming 60 minutes of AI-generated music in the same key and tempo, and ships a game with a real soundtrack instead of a Kevin MacLeod placeholder.

The next step is opening Music Gen in a tab, picking a level concept from the project, writing a six-part prompt (genre, mood, instrumentation, tempo, structure, negatives), and pressing Generate. The 100 starter credits cover the first 10 tracks, the SFX pass costs another 60 credits, and the in-engine loop trim happens in SFX Editor without ever leaving the browser. The full path is at Sorceress Sound Studio; the tool catalog at /tools-guide covers the rest of the asset stack (sprites, tilesets, 3D models, voice) that completes the soundtrack’s home. The composer step is no longer the blocker. The next blocker is the level itself.