Game Audio

Use this reference for music, SFX, ambience, and spoken character lines. Generate through Higgsfield CLI, then normalize and mix locally for predictable in-game loudness.

Discover the live contracts

Run before the first audio job:

higgsfield model list --audio --json
higgsfield model get seed_audio

Use seed_audio by default. Use sonilo_music or mirelo_text_to_audio only when the user requests those specialist models or their live contracts fit better. Use inworld_text_to_speech only for explicit TTS with one of its listed voices.

Commands

General audio, SFX, or ambience:

higgsfield generate create seed_audio \
  --prompt "short isolated sword impact, dry studio recording, no music" \
  --wait \
  --json

Specialist instrumental music:

higgsfield model get sonilo_music
higgsfield generate create sonilo_music \
  --prompt "instrumental cozy forest loop, warm marimba and soft strings, no vocals" \
  --duration 30 \
  --wait \
  --json

Specialist legacy SFX:

higgsfield model get mirelo_text_to_audio
higgsfield generate create mirelo_text_to_audio \
  --prompt "single heavy wooden door slam, close microphone, no ambience" \
  --duration 2 \
  --wait \
  --json

Explicit TTS:

higgsfield model get inworld_text_to_speech
higgsfield generate create inworld_text_to_speech \
  --prompt "The gate is open. Move!" \
  --voice "<exact voice value from model get>" \
  --wait \
  --json

Pass only parameters shown by model get; do not forward provider-specific keys.

Scope

• Music: at most two looping tracks for a small game; one per major scene is usually enough.
• SFX: prioritize the primary verb, damage/feedback, pickup/reward, important environment, then one ambience layer. Keep small games to roughly five essential effects.
• Voice: lock one voice per speaking entity and reuse it. Keep lines short enough not to block gameplay.
• Generate independent clips concurrently, but wait for every manifest row and save its result under the project assets/ directory.

Prompt rules

• Describe one sound per SFX prompt. Do not request a complete mixed scene.
• State no music, no voice, or no ambience when isolation matters.
• Music prompts specify mood, tempo, instrumentation, and instrumental/no vocals when lyrics are unwanted.
• Voice prompts contain the exact line; performance direction belongs in the model-supported prompt, not undocumented flags.
• Preserve the game's STYLE FORMULA conceptually through material, era, energy, and tonal language even though audio is not visual.

Mix and ear safety

Normalize locally before wiring clips into the game:

• Voice around -6 dBFS.
• SFX around -10 to -12 dBFS.
• Music around -18 to -20 dBFS.
• Final true peak at or below -3 dBFS.

Voice stays above SFX; SFX stays above music. Loop ambience/music in playback code and add short fades at loop boundaries. Never stack raw model outputs at full gain.

Verification

• Every audio manifest row resolves without a 404.
• Autoplay restrictions are handled: start/resume the audio context from a user gesture.
• Pause or mute audio when the page loses focus if that matches the game profile.
• The game remains playable with sound disabled.
• Test on phone speakers as well as headphones when mobile is in scope.