ElevenLabs SFX v3: sound effects and full scenes from one prompt
Describe a sound or a whole scene and ElevenLabs SFX v3 returns an original, royalty-free sound effect up to 30 seconds long, with optional looping. Its newest step over SFX v2 is following multi-sound prompts in the order you write them.
ElevenLabs SFX v3 is ElevenLabs' newest sound-effects model: describe a sound or a whole scene in text and it returns an original, royalty-free clip up to 30 seconds long. Name several sounds in one prompt and it renders them together, in the order you wrote. The biggest step over SFX v2 is reliability — more of the sounds you list actually land. It also handles short music and vocals better, and keeps resonant sounds clean.
ModelElevenLabs SFX v3Generate layered sound effects from one text prompt.What it does well
Several sounds in one prompt, in order
SFX v3 is built to follow prompts that name more than one sound. Describe each event in the order it should play, and the model renders them together as one clip. In testing, a forge prompt produced three distinct, time-ordered events in an 8-second clip: a hammer strike, the steel ringing and fading, then a hiss of steam.
A blacksmith's hammer strikes hot steel on an anvil, the metal ringing and fading between blows, then the glowing blade plunges into a water trough with a sharp hiss of steam
Listen to the output — three named events render in order as one 8-second clip; list each sound in the order you want to hear it.
Music and vocals, not just noises
SFX v3 also handles short musical phrases and vocal textures. You can pair an instrument and a voice with an environment, and the parts stay distinct rather than garbling together. It is built for stings, accents and loops, not full songs.
A street busker fingerpicks a phrase on a steel-string guitar while a woman sings a short wordless melody over it; footsteps and murmurs echo through a subway station
Listen to the output — a pitched guitar phrase and a sung melody sit over subway ambience without garbling; name the instrument, the voice and the surroundings together.
Clean, resonant high-frequency sounds
Chimes, gongs and whistles can turn harsh or crackly when generated. SFX v3 holds their tone and decay without that clipping. Describe the arc of the sound — rise, hold, release — and the model tracks it.
A stovetop kettle heats on the burner, its rising hiss building into a steady high whistle, then a hand lifts it clear and the whistle falls away
Listen to the output — the kettle's high whistle holds a steady, clean tone then falls as it is lifted, with no harsh clipping; describe the arc: rise, hold, release.
Examples to try
For ambient beds, turn looping on and push the duration toward 20 to 30 seconds so the clip has room to settle.
A tropical coral reef underwater - steady water movement, the constant crackle of snapping shrimp, and a distant whale call rolling past
Listen to the output — a 30-second underwater reef bed repeats with no audible seam; turn loop on and push duration toward 20 to 30 seconds for beds.
More prompts worth trying: "Footsteps on gravel, then a metal gate opens, then wind through trees", "Vintage brass stabs in F minor", and "Heavy rain on a tin roof, thunder rolling in the distance".
Recommended settings
| Setting | Value | When to use it |
|---|---|---|
| Prompt | Action, setting and atmosphere, in order | Always; sequence the events for layered effects |
| duration_seconds | 0.5 to 30 (8 for layered scenes) | Set it to fit a cut; omit it to let the model pick a length |
| loop | on | Ambient beds and backgrounds that must repeat |
| prompt_influence | 0 to 1 (0.5 in testing) | Raise toward 0.7 to 1.0 for a literal brief; lower toward 0.1 to 0.3 for variety |
| output_format | mp3_44100_128 or wav_48000 | MP3 for general use; WAV 48 kHz for high-quality masters |
Good to know
- Clips run up to 30 seconds per generation; for full songs, reach for a dedicated music model.
- Exact repetition counts, such as "triangle hit four times", are not always followed.
- Small wording changes shift the result, so try a synonym if a sound comes out wrong.
Use it from the API
Send a prompt and the settings you want, and the model returns an audio file URL. The default example:
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"elevenlabs-sfx-v3",
prompt="A blacksmith's hammer strikes hot steel on an anvil, the metal ringing and fading between blows, then the glowing blade plunges into a water trough with a sharp hiss of steam",
duration_seconds=8,
prompt_influence=0.5,
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/elevenlabs-sfx-v3" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A blacksmith'\''s hammer strikes hot steel on an anvil, the metal ringing and fading between blows, then the glowing blade plunges into a water trough with a sharp hiss of steam",
"duration_seconds": 8,
"prompt_influence": 0.5
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Resources
- Try ElevenLabs SFX v3 on Segmind
- API reference
- Sound effects | ElevenLabs Documentation — how to prompt SFX v3, with parameters and key limits.
- Introducing SFX v3 in ElevenCreative and ElevenAPI — what SFX v3 improves over v2.
- Text to Sound Effects — API reference — the sound-effects convert endpoint and its parameters.
- Models | ElevenLabs Documentation — ElevenLabs model overview and model IDs.
- #text-to-audio
- #elevenlabs
- #model-launch
- #sound-effects
- #sound-design
