Trains mark a period. A steam locomotive, a diesel express and a modern electric metro each place a scene in a different decade before any dialogue, and audiences read them immediately. Trains are also large, so perspective matters more than usual: trackside, on the platform and inside the carriage are three different sounds from the same train.
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.
steam locomotive departing a station, slow chuffing accelerating with wheel slip and a whistle, trackside, open air
steam locomotive whistle blowing long and mournful, multi-tone brass with a valley echo, distant outdoor
modern electric train passing through a station without stopping, rising roar and doppler, platform perspective
inside a subway carriage between stations, rhythmic rail clatter with motor whine and rattling panels, seamless loop
railway level crossing bells ringing and barriers lowering, roadside, outdoor, traffic waiting
long freight train rolling slowly past, endless wheel clatter and couplings knocking, trackside, distant horn
train braking into a station, steel wheels squealing sharply and grinding to a stop, platform reverb
railway station concourse ambience, echoing announcements, footsteps and distant train movement, large hall
modern train doors beeping then hissing closed with a pneumatic thump, platform, close
steady rhythmic clatter of wheels over rail joints at speed, trackside close, seamless loop
These models respond to physical description. Name the material, the action and the space.
steam locomotive departing a station, slow chuffing accelerating with wheel slip and a whistle, trackside, open air
"Wheel slip" makes a steam departure believable. The moment of lost traction is in the real recordings and missing from generic prompts.
modern electric train passing through a station without stopping, rising roar and doppler, platform perspective
"Without stopping" implies speed, and "platform perspective" puts the listener beside the track.
inside a subway carriage between stations, rhythmic rail clatter with motor whine and rattling panels, seamless loop
"Between stations" keeps the bed steady. Otherwise the model adds a stop, which breaks the loop.
steady rhythmic clatter of wheels over rail joints at speed, trackside close, seamless loop
Isolating the rail joints gives you a rhythmic layer that works under any train, which makes it more reusable than a whole train.
Steam chuffs and whistles; diesel clatters and sounds a horn; electric hums and whines. That choice places a scene in a decade more reliably than set dressing, and the models handle it well because the three families are distinct in the training data. State it: "steam locomotive", "diesel locomotive", "modern electric train". Plain "train" leaves the choice to the model.
The classic train rhythm comes from wheels crossing gaps in the rail, which is why modern continuously welded track sounds much smoother than older jointed track. For the clackety rhythm, ask for "rail joints". For a modern high-speed line, ask for a smooth continuous roar. Each fits a different decade.
A station is a large hard enclosure with a long reverb tail and echoing announcements. Open track has almost no reflection beyond ground bounce. Moving the same train between them changes it completely, and it takes one word in the prompt. When a train sound feels wrong for a shot, check the space first. Most mismatches are a trackside recording under a platform shot, or the reverse.
The models run on your machine, which changes the cost, the privacy and what you can automate.
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.
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.
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.
claude mcp add --transport http soundbox http://127.0.0.1:17494/mcpcurl -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}'