Introduction

ElevenLabs just released SFX v3, the newest version of its text-to-sound-effects model, and the pitch is simple: describe a whole scene, not one sound. A football clip that needs a crowd cheering, a ball dropping and the same crowd booing a few seconds later used to take a pile of retries. SFX v3 is built to land all three from a single prompt.

It's closed and hosted, announced October 29, and you can use it today in ElevenCreative (the Sound Effects tab, Flows, Studio and the Music Editor), in the ElevenLabs mobile app, and through the API with model_id set to eleven_text_to_sound_v3. One catch up front: the docs call it limited access, possibly not in every region, and accounts without it get an Unknown model error.

What Changed From SFX v2

From the launch post and the sound effects docs:

  • Multi-sound prompts: v3 follows several instructions inside one prompt more reliably, so more of the sounds you list actually show up in the clip.
  • Music and vocals: prompts that mix a musical phrase or a vocal performance with everyday sounds come out more convincing than on v2.
  • Cleaner high end: fewer crackles and clipping on bright sounds like gongs, chimes and whistles, so a gong keeps its resonance and a whistle holds its tone.
  • Same controls: up to 30 seconds per generation, seamless looping for ambience, a prompt-influence slider, MP3 for everything and 48 kHz WAV for non-looping effects.

ElevenLabs says v3 beat v2 in its own benchmark testing but hasn't published the scores, and the docs themselves say results vary by prompt. So treat it as "probably better, check with your ears," not a guaranteed upgrade.

Pricing

  • Credits: the docs list 40 credits per second when you set a duration.
  • API rate: ElevenLabs' API pricing page lists Sound Effects at $0.12 per minute of generated audio, royalty-free.

ElevenLabs hasn't announced a separate price for v3, so expect the same Sound Effects rate as before.

How to Try It Today

  1. Open the Sound Effects tab in ElevenCreative and pick SFX v3 as the model.
  2. Write the scene in order: the main action first, then the surroundings, then the mood.
  3. Generate the same prompt on v2 and v3 and keep whichever sits better under your video.

If you'd rather script it, this is the request body for POST /v1/sound-generation:

{
  "model_id": "eleven_text_to_sound_v3",
  "text": "Football crowd cheering, a ball drops on the pitch, then the crowd boos",
  "duration_seconds": 10,
  "prompt_influence": 0.5
}

Pro Tip

Describe actions and places, not categories. "Footsteps on gravel approaching slowly" beats "footsteps," and ElevenLabs' own example found that "a couple arguing in French" worked better than "a couple fighting in French." Don't count on exact repetitions, though; "triangle hit 4 times" isn't always followed, so trim counts in your editor instead.

Also From ElevenLabs This Week: Cue

A day earlier, ElevenLabs shared details on Cue, the forced alignment model that lines a transcript up with its audio, character by character. It already powers every Scribe transcription and the Forced Alignment API, and ElevenLabs reports it aligns an hour of audio in about half a second on one GPU. The fun part for music people: on a set of Creative Commons songs it placed 92.7% of sung words within 0.3 seconds, versus 82.0% for the next best tool, which is exactly what you want for lyric videos and karaoke timing.

Why It Matters for Audio Creators

Sound design is still the most hands-on part of AI video. If one prompt can sketch a scene's whole soundbed, with action, crowd and a musical sting together, you can rough in the audio for a clip in minutes and spend your time fixing the moments that matter. Pair it with Cue for subtitles or lyric timing and most of the audio side of a short now lives in one place.

Conclusion

SFX v3 is ElevenLabs' new closed sound effects model: better at multi-sound scenes, music and vocals, live in ElevenCreative and the API as eleven_text_to_sound_v3, and still in limited access. Run your favorite v2 prompts through it side by side and let your ears decide.

—Anabel ♡