Blog

Guide

ElevenLabs Sound Effects v3 API: 30-Second Sound Effects, Seamless Loops and Moving from v2

Pricing, parameters, example requests and prompting tips in one guide

Nov 11, 2026 6 min read
Faceted low-poly mineral cross-section on a deep ink-green background with teal and malachite tones
Model
elevenlabs-sound-effects
On this page

What Sound Effects v3 does

ElevenLabs Sound Effects v3 is a text-to-sound-effect model on HiAPI, with the model ID elevenlabs/sound-effects-v3. Describe the sound you want, such as the action, the setting, the atmosphere and how it should be performed, and the model generates a matching sound effect. Clips can be 0.5 to 30 seconds long, and the model can also generate loops that play seamlessly on repeat.

Each call returns one MP3 clip. This guide covers pricing, parameters, example requests, how v3 differs from v2, and how to write prompts. All numbers reflect the API definition at the time of writing.

Sound Effects v3 API pricing

You pay $0.0029 per second of requested duration (duration_seconds).

  • 2 seconds (the default): $0.0058
  • 5 seconds: $0.0145
  • 10 seconds: $0.029
  • 30 seconds (the maximum): $0.087

Turning on loop or changing prompt_influence costs nothing extra. Failed tasks are refunded in full. v2 also costs $0.0029 per second.

Parameters

Only prompt and duration_seconds are required.

  • prompt: required, 1–450 characters. Describe the sound you want.
  • duration_seconds: required, 0.5–30 seconds. There is no automatic duration.
  • loop: generate a sound effect that loops seamlessly. Defaults to false.
  • prompt_influence: 0–1, defaults to 0.3. Higher values follow the prompt more literally; lower values allow more variation.

Making a request

Sound generation on HiAPI is asynchronous: you create a task, then poll for the result. model goes at the top level and everything else goes inside input.

A 5-second door sound:

curl -X POST https://api.hiapi.ai/v1/tasks \
  -H "Authorization: Bearer $HIAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "elevenlabs/sound-effects-v3",
    "input": {
      "prompt": "Heavy wooden door creaking open slowly in a quiet stone hallway, then a soft thud as it stops",
      "duration_seconds": 5
    }
  }'

A 10-second rain loop:

{
  "model": "elevenlabs/sound-effects-v3",
  "input": {
    "prompt": "Gentle rain on a tin roof with distant rolling thunder, calm night ambience, no music",
    "duration_seconds": 10,
    "loop": true
  }
}

The second request costs 10 × $0.0029 = $0.029. Once the task is created, poll GET https://api.hiapi.ai/v1/tasks/{taskId} until it succeeds, then download the audio URL from the output.

How v3 differs from v2

Both versions use the same parameter names and the same price. On HiAPI, they differ in three ways:

  • Maximum duration: up to 30 seconds for v3 and 22 seconds for v2. For ambience or background beds between 22 and 30 seconds, use v3.
  • Output quality: ElevenLabs' documentation says v3 follows prompts with several instructions more closely, handles prompts with music or vocals better, sounds more natural, and is less likely to clip on high-frequency sounds such as chimes and whistles. It also notes that results vary by prompt and recommends comparing both models on your own prompts.
  • Lifecycle: ElevenLabs has deprecated v2 (eleven_text_to_sound_v2) and plans to remove it on March 1, 2027. Use v3 for new projects.

To move from v2 to v3, change model to elevenlabs/sound-effects-v3. Nothing else needs to change. Before switching production traffic, run a few of your usual prompts through both versions and compare the results.

Writing prompts

  • Describe the action, setting, atmosphere and performance rather than naming a broad category. "Slow footsteps on wet gravel" gives more consistent results than "footsteps".
  • Describe a sequence when you need one, for example "Footsteps on gravel, then a metallic door opens".
  • Use sound-design terms: impact, whoosh, ambience, one-shot, drone and so on.
  • Don't rely on exact repetition counts. ElevenLabs notes that prompts such as "triangle hit 4 times" are not always followed. If you need a precise rhythm, generate once and edit the clip yourself.
  • Tune prompt_influence: for UI sounds and notifications that must sound the same every time, try 0.6 or higher. To explore a range of options, keep the default 0.3 or go lower.

Current limits

  • Each call returns one clip. For several candidates, send several tasks in parallel.
  • Duration is 0.5–30 seconds and must be set explicitly. There is no automatic duration.
  • Prompts are limited to 450 characters.
  • Only asynchronous tasks are available. There is no synchronous response.
  • Output is MP3. Other formats are not available yet.

Where it fits

v3 works well when you need many short sound effects quickly: UI and notification sounds for games and apps, transitions for short videos and ads, temp foley for film and trailer previs, and looping ambience for audiobooks and live streams. A 5-second clip costs $0.0145, so you can generate several versions, keep the best one, and only then take it into post-production. That is usually cheaper than repeatedly searching for and editing stock sounds.

Related articles

View all articles →

Start building with persistent AI artifacts

Get your API key and call image, video and audio models through one task-based API.

Read the API docs