Sound designers use glass constantly and often get it wrong, because "glass breaking" covers at least six different sounds: a thin wine glass ringing itself apart, a bottle bursting on concrete, a car window collapsing into pebbles, a pane failing all at once. Thickness matters most, then what the glass lands on. Soundbox runs the models on your own GPU, so you can render all six, compare them against the picture, and regenerate the one that nearly worked.
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.
glass bottle smashing on concrete, sharp initial burst with shards scattering and skittering away, outdoor alley reverb
large window pane breaking, cracking then collapsing inward with long shards falling and shattering on a wooden floor
wet finger circling the rim of a wine glass, pure sustained crystalline tone slowly building, quiet room
two wine glasses clinking together in a toast, bright ringing chime with a short decay, close mic
car side window shattering into thousands of small tempered pebbles, dense granular collapse onto a road
glass cracking slowly under pressure, fine splintering fractures spreading with tiny ticks, very close mic, quiet
glass jar dropped onto a tiled kitchen floor, hard burst with a wet contents splash and shards spinning
lightbulb filament popping and the thin glass envelope bursting, delicate high shatter, close and dry
broom sweeping broken glass across a tile floor, continuous granular scraping with shards tumbling
greenhouse roof collapsing, sequence of large panes failing one after another with a long debris tail
These models respond to physical description. Name the material, the action and the space.
glass bottle smashing on concrete, sharp initial burst with shards scattering and skittering away, outdoor alley reverb
The surface it hits and the aftermath do the work. "Skittering away" turns a single hit into a break.
car side window shattering into thousands of small tempered pebbles, dense granular collapse onto a road
Tempered glass breaks into small pebbles instead of shards. "Pebbles" and "granular" are what separate a car window from a house window.
wet finger circling the rim of a wine glass, pure sustained crystalline tone slowly building, quiet room
Describe the physical action. The model breaks glass by default as soon as the word appears, and naming the finger on the rim keeps it intact.
glass cracking slowly under pressure, fine splintering fractures spreading with tiny ticks, very close mic, quiet
"Very close mic" and "quiet" keep the small details audible. A fracture is a tiny sound and any room tone buries it.
Every glass sound depends on how thick the glass is. Thin glass (bulbs, wine glasses, picture frames) breaks bright, fast and high, with almost no low end. Thick glass (shopfronts, greenhouse panes, bottles) has real body and a lower initial crack before the debris starts. If a generation sounds like the wrong object, more adjectives about breaking rarely help. Name the object precisely enough that its thickness is implied.
About half of a glass break is the surface receiving the debris. Shards on concrete skitter and ring; on carpet they thud and stop; on tile they bounce twice and scatter wide; in a sink they clatter metallically. If you leave the surface out, the model picks one, usually a vague dry nothing. One extra clause, such as "onto a wooden floor" or "into a metal sink", changes the result more than any tuning.
A convincing break has an impact, a failure and a settle. Cheap library glass gives you the first two and cuts. If the clip has to sit under a shot without a second layer, write the settle into the prompt ("shards scattering and skittering away", "debris tail", "small pieces still falling") and give the render enough seconds to reach it. Two seconds holds an impact; a full break needs about four.
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}'