Metal sound effects you generate yourself

Metal is the most tonal material in a sound library. Struck wood or stone stops quickly; struck metal keeps ringing, and the ring tells you the size of the object for as long as it lasts. That makes metal the raw material for much of horror and most industrial sci-fi. These prompts treat resonance as the main thing you specify.

Listen: 10 samples from the metal collection

Every clip shows the exact prompt that produced it. The render script reads its prompts from the same file as this page. Copy one, change a word and generate your own.

  • Sheet wobble

    via audiocpp-stable-audio-3-sfx

    large thin metal sheet being flexed and wobbled, deep warbling thunder-like resonance, close mic, dry

  • Pipe strike

    via audiocpp-stable-audio-3-sfx

    steel pipe struck hard with a wrench, bright metallic clang with a long ringing sustain, empty warehouse

  • Chain drag

    via audiocpp-stable-audio-3-sfx

    heavy iron chain dragged slowly across a concrete floor, dense clinking links, basement reverb

  • Hull groan

    via audiocpp-stable-audio-3-sfx

    ship hull groaning under pressure, low metallic strain with creaking rivets, enormous enclosed space

  • Rust scrape

    via audiocpp-stable-audio-3-sfx

    rusted metal scraping against rusted metal, harsh grinding rasp with a squealing high edge, close mic

  • Forge hammering

    via audiocpp-stable-audio-3-sfx

    blacksmith hammering hot steel on an anvil, rhythmic bright strikes with a ringing tail, stone forge

  • Can crush

    via audiocpp-stable-audio-3-sfx

    aluminium can crushed in a fist, thin buckling metal with a crinkling collapse, very close mic, dry

  • Girder drop

    via audiocpp-stable-audio-3-sfx

    steel girder dropped onto concrete, enormous low clang with a slow decaying wobble, construction site

  • Wire twang

    via audiocpp-stable-audio-3-sfx

    taut steel wire plucked hard, sharp descending twang with a long metallic decay, close and dry

  • Shutter roll

    via audiocpp-stable-audio-3-sfx

    corrugated metal shop shutter rolled down fast and slamming, rattling ribs with a hard final clang, street

Prompts that work

These models respond to physical description. Name the material, the action and the space.

  • Thin and tonal

    large thin metal sheet being flexed and wobbled, deep warbling thunder-like resonance, close mic, dry

    Thin metal bends instead of ringing. "Warbling" gets you the bend; without it you get a clang.

  • Thick and ringing

    steel pipe struck hard with a wrench, bright metallic clang with a long ringing sustain, empty warehouse

    Naming the striker as well as the object fixes the transient. The room then supplies the sustain.

  • Metal under load

    ship hull groaning under pressure, low metallic strain with creaking rivets, enormous enclosed space

    "Under pressure" turns an impact into a groan: it tells the model the force is continuous.

  • Friction, not impact

    rusted metal scraping against rusted metal, harsh grinding rasp with a squealing high edge, close mic

    Saying both surfaces are rusted doubles the texture. A scrape between two named materials is more specific than "metal scrape".

Making metal sounds that fit

Resonance length is the size control

You judge the size of a piece of metal by how long it rings after the hit more than by the attack. A small object rings for a fraction of a second; a girder or a hull rings for many seconds and wobbles as different modes decay at different rates. When a metal sound reads as the wrong size, change the tail and leave the hit alone: ask for "a long ringing sustain" or "dead, no ring", and give the render enough seconds for the tail to fit.

Rust is a texture word that works

Clean and rusted metal sound different: clean surfaces slide and ring, corroded ones catch, rasp and squeal. "Rusted", "corroded" and "old" are among the few adjectives these models respond to reliably, because they map onto real spectral differences. If a metal scrape comes back too smooth and synthetic, add rust before you rewrite the action.

Metal for horror

Many unsettling horror sounds are metal recorded in unusual ways: a sheet flexed slowly, a wire bowed, a hull straining. Metal produces inharmonic partials, which sound tonal without fitting any musical scale, and that makes a sound feel wrong. If you are building dread, generate the source here and pitch it down. You get the same kind of result as the horror collection, with more control over the source.

Why generate locally

The models run on your machine, which changes the cost, the privacy and what you can automate.

Offline after download

Weights are fetched once. After that, generation, the library and the API all work with the network off. Everything lives in one SQLite file and one folder of WAVs under ~/.soundbox, and nothing is uploaded.

Your agent can use it

Soundbox runs an MCP server in the same process, with tools for generating and listing sounds. Generated files come back as absolute paths, so an agent can copy them into a project.

Licences stated up front

Every model's terms are shown before you download it, and non-commercial weights need an explicit confirmation. Soundbox is MIT. The licence of the model that made a sound governs what you may do with it.

Connect a coding agent
claude mcp add --transport http soundbox http://127.0.0.1:17494/mcp
Generate over HTTP
curl -X POST http://127.0.0.1:17494/api/generate \
  -H 'Content-Type: application/json' \
  -d '{"model":"audiocpp-stable-audio-3-sfx","prompt":"8-bit coin pickup blip","duration_s":1}'

Frequently asked questions

How do I stop every metal hit sounding like the same clang?
Vary the striker and the space independently. "Struck with a wrench", "struck with a rubber mallet" and "struck with a bare hand" give different transients from the same object, and moving the same hit from "close mic, dry" to "empty warehouse" changes the tail completely. Sample packs sound alike because every clip was recorded in one room with one striker; generating lets you change both.
Can I use these sounds in a commercial video or game?
It depends on the model that made them. Soundbox shows each model's licence before you download its weights. The clips on this page were rendered with Stable Audio 3.0 Small SFX, which uses the Stability AI Community License: commercial use is free below USD $1M annual revenue, and above that Stability requires an enterprise licence. Soundbox itself is MIT, but the audio is governed by the licence of the model that made it. Read that licence before you ship; this summary is not legal advice.
Does this work offline?
Yes, once the weights are on disk. Generation, the library and the HTTP API all run locally against one SQLite file and one folder of WAVs under ~/.soundbox. There is no account and no API key, and nothing is uploaded. The only step that uses the network is the initial model download.
What hardware does it need?
The clips here were rendered on a consumer NVIDIA GPU at roughly ten seconds of wall time per clip, through audio.cpp's quantised GGUF runtime, which needs little VRAM. CPU-only generation is supported, but expect minutes per clip instead of seconds: fine for a handful of sounds, slow for a whole library.
Can an agent generate these for me?
Yes. While the app is open, Soundbox runs an MCP server at http://127.0.0.1:17494/mcp with tools for generating, listing and mixing sounds. Every generated file comes back as an absolute path, so a coding agent can render the sound a scene needs and put it straight into the project. The several hundred clips across these pages were made this way.