Procedural SFX & Scene Composer

Generate fresh, sample-free sound effects for UI prototypes, simple games, and motion graphics. Choose individual sound families for a WAV pack, or combine supported cues into a short scene. The sound is synthesized mathematically inside the Actor: no external audio-generation API, downloaded samples, or speech model is used.

What you get

• Sound pack: 1–12 sound families, with 1–3 structurally different variants each (up to 36 WAV files). Available families: ui_click, ui_confirm, ui_error, ui_notification, coin, jump, laser, hit, explosion, whoosh, bell, and ambient_drone.
• Short scene: a 2–12 second mixed WAV with 1–8 supported cues, plus a cue sheet and individual cue stems in a ZIP. You can provide a short English/Russian description or an explicit timeline.
• Inspectable output: SFX_PACK.zip, MANIFEST JSON, OUTPUT JSON, and a Dataset row per generated sound or scene cue. In scene mode, SCENE.wav is also available directly.

All audio is 44.1 kHz, mono, 16-bit PCM WAV. The same seed and settings reproduce byte-identical output within the same engine/runtime build; later engine versions may change the result. Preserve the manifest and hashes if exact reproducibility matters.

Quick start

Generate all 12 families with three variants each (the default input):

{"mode":"pack","seed":"my-ui-pack","variants":3}

Generate only selected sounds:

{"mode":"pack","seed":"game-01","presets":["coin","jump","laser","hit","explosion"],"variants":2}

Compose a short scene from recognized words:

{"mode":"scene","seed":"demo-scene-01","sceneDurationSeconds":5,"sceneDescription":"Low drone, then a notification, laser and explosion"}

For predictable timing, an agent or developer can supply an explicit plan instead:

{
  "mode":"scene",
  "seed":"demo-scene-01",
  "scenePlan":{
    "durationSeconds":5,
    "cues":[
      {"preset":"ambient_drone","variant":1,"atSeconds":0,"gainDb":-11,"loop":true},
      {"preset":"ui_notification","variant":2,"atSeconds":1,"gainDb":-5},
      {"preset":"explosion","variant":3,"atSeconds":3,"gainDb":-6}
    ]
  }
}

Finding your files

Open the run's Output tab to download the ZIP or read OUTPUT. The ZIP contains every WAV in pack mode. In scene mode it contains scene.wav, cue-sheet.json, and a stems/ folder. The Dataset contains metadata for each generated item; it is not the binary audio itself. The key-value store holds SFX_PACK.zip, MANIFEST, OUTPUT, and in scene mode SCENE.wav.

For automated workflows, inspect OUTPUT.status, mode, the file keys, and Dataset rows. In scene mode, inspect unsupportedCues and unparsedWords before treating the result as complete. Use the manifest's SHA-256 hashes when you need to verify a download.

Important limits

This is a controlled procedural synth, not general text-to-audio or realistic Foley. Free-text scene planning recognizes only the 12 listed sound families. It follows their order in the description and spreads foreground sounds across the duration; words such as “then” do not define exact timestamps. It cannot synthesize dialogue, animal sounds, rain, wind, footsteps, doors, vehicles, or music. Known unsupported concepts and other unparsed words are reported, and OUTPUT.status becomes partial if at least one supported cue was rendered. A description with no supported cue is rejected instead of producing a misleading file. Use scenePlan for precise cue placement, but it still supports only the same 12 families.

The current output is intentionally mono WAV, not MP3 or stereo. Up to eight cues and twelve seconds keep runs bounded. The Actor does not upload audio to any third-party sound service. Apify still stores the generated files as part of the run and charges for platform usage under your Apify plan.