ElevenLabs Sound Effects V2: foley and looping ambience from text

A door that creaks in exactly the right way. Rain on a tin roof that runs under a whole scene without a seam. A spaceship that hums, crackles and then tears a hole in space. Finding these in a sound library means an afternoon of scrolling; ElevenLabs Sound Effects V2 writes them from a sentence. You describe the sound, set how long it lasts, and get an audio file back. It is one of the cheapest models on the models page, which makes it the right place to experiment.

Think like a foley artist, not a poet

The prompt box asks for three things, and it is worth answering all of them: the sound itself, the place it happens in and how it changes over time. "A door" gives you a generic door. "A heavy wooden door in a stone hallway, slow creak, then a deep thud and a short echo" gives you a scene. The prompt has a character limit of a short paragraph, so spend the words on the source, the space and the order of events, not on adjectives like "amazing".

Look at the example on the model page: a starship builds a reactor hum, then electric arcs, then a metallic roar as space tears, then a whoosh, then echoes fading to silence. That is a timeline, written in the order you will hear it, and it is why the result sounds designed rather than random.

The four settings

• Prompt. Required. The description above.
• Duration. A slider in seconds, from a short blip to a stretch long enough for a slow build and a fade. The default is a few seconds, which suits most single effects.
• Loop (advanced). Asks for a sound designed to repeat smoothly. Switch it on for ambience: rain, a crowd, a server room, wind in trees.
• Prompt influence (advanced). Higher values follow your words more closely; lower values give the model more freedom. The default sits low.

Three prompts to start from

Lifehacks

• Say what you do not want. The example prompt ends with "Sound effects only, no speech, no music". Borrow that line whenever a stray melody or murmur would spoil the take.
• Draft short, then stretch. Try the idea at a short duration first. When the character of the sound is right, run it again at the full length you need.
• Raise prompt influence for precise effects. For a specific sequence (a click, then a beep, then a hiss) push the slider up. For a mood like "eerie forest at night", leave it low and let the model fill the space.
• Layer two simple sounds in your editor instead of asking for one complicated one. A wind bed plus a separate creaking sign mixes better than both in one prompt.

Where it stops

It hears only your words. There is no video input, so it cannot watch a clip and hit the moment a glass breaks; you line the sound up yourself in an editor. One run is also capped at a modest length, so a long ambience is a loop repeated, not one continuous recording. It makes effects and ambience, not a clean voice line or a finished song: for speech, use a speech model, and to clean up a recording you already have, ElevenLabs Audio Isolation is the neighbour you want.