ElevenLabs sound effects: SFX v3 builds full scenes from one prompt

ElevenLabs has released SFX v3, the third version of its AI sound effects model. The company says the new ElevenLabs sound effects model builds richer, more precise sounds from a single text prompt and handles music and vocals better. SFX v3 is available in ElevenCreative and through the sound effects API.

What SFX v3 changes

ElevenLabs pitches SFX v3 as a model for whole scenes rather than single sounds. "Big scenes rarely come down to one sound," the company wrote when it announced the model on X on October 29. The idea is that one prompt can now carry several layers at once, such as ambience, action and music.

The launch video shows how that works. One prompt describes two people arguing in Italian, a slammed door, footsteps, a hit table, breaking glass and a slap. The model returns them as one sound scene, which ElevenLabs sums up as "one prompt, a complete sound scene."

The second half of the video covers short prompts. They include rain on a car roof, footsteps above an old wooden ceiling and a distant train in fog. It ends in ElevenCreative, where a menu switches the sound effects model between V2 and V3.

ElevenLabs makes three claims about the new version:

• richer sound effects from one text prompt, with more layers in a single generation;
• more precise results that follow the prompt more closely;
• better music and vocal elements inside a sound effect.

All three are the company's own descriptions. ElevenLabs has published no listening tests or comparisons with SFX v2, so the size of the jump is unknown. What the API documentation does show is how developers reach the new model.

How to use SFX v3 in the sound effects API

The text to sound effects endpoint now lists two model IDs, eleven_text_to_sound_v2 and eleven_text_to_sound_v3. The default is still v2, so an API call gets the old model unless it names v3 explicitly.

The documented limits have not changed with the new ID:

• duration from 0.5 to 30 seconds, or set automatically from the prompt when left empty;
• a loop option for sounds that repeat without an audible start or end;
• prompt_influence from 0 to 1, with 0.3 as the default;
• output as MP3, Opus, PCM, WAV or the ulaw and alaw telephony formats.

Some formats depend on the plan. MP3 at 192 kbps needs the Creator tier, and 44.1 kHz PCM or WAV needs the Pro tier. The sound effects guide lists 40 credits per second when a duration is set, but it does not say whether SFX v3 costs the same. Those limits look the same as the ones SFX v2 brought a year ago.

From single sounds to sound scenes

ElevenLabs launched its first sound effects model on May 31, 2024. It produced short one-off sounds from a text description, such as a door slam or footsteps.

SFX v2 followed on September 2, 2025. According to The Decoder, it raised the clip length to 30 seconds and added seamless looping. It also moved to 48 kHz and expanded the sound library.
SFX v3 keeps the 30-second ceiling and the looping option. The change ElevenLabs describes is about what goes into those 30 seconds, not how long they last. The model also lands at the end of a busy release run for the company.

Where SFX v3 fits in ElevenLabs' fall releases

SFX v3 is the latest in a string of ElevenLabs launches this fall. The company has been building one place to make video, voice, music and sound together.

The main steps before SFX v3:

• September 21: Studio 4.0, an editor for video, images, speech, music and sound effects in ElevenCreative;
• September 28: Eleven v4 and Eleven v4 Turbo, the new text-to-speech models;
• October 15: Video to Music, which generates a soundtrack from one or more videos;
• October 19: Zero Retention Mode for enterprise customers and new WAV formats on the sound effects endpoint;
• October 20: Eleven Realtime, speech-to-speech models for voice agents.

Better music and vocals inside a sound effect blur the line between SFX v3 and the company's music and speech models. That makes the sound effects model one more piece of the same editor. ElevenLabs is not the only company chasing that editor, though.

Rivals among AI sound effects generators

The competition comes from both creative software makers and model labs. Adobe added its own AI sound effects generator, Generate Sound Effects, to Firefly on August 20, 2026, as AlternativeTo reported. It runs on the Firefly Audio Model and matches the timing and energy of the user's content.

Stability AI released Stable Audio 3.0 on May 20, 2026, as a family of four models. eWEEK, citing TechCrunch, reports that the line-up includes a 459 million-parameter Small SFX model for sound effects on phones and laptops. Stability says every Stable Audio 3.0 model was trained on fully licensed data.

Video generators add pressure from a third side. Models such as Google's Veo already put sound into the clips they generate. That leaves a standalone model like SFX v3 to win on control and quality, and those are hard to judge from the announcement alone.

So is SFX v3 a real step up? For now it is a promise more than a proven upgrade. The announcement gives no numbers, no pricing for the new model and no comparison with v2. In the API, SFX v3 does not even replace v2 as the default.

What has changed is the job ElevenLabs sets for the model. SFX v1 made single sounds, SFX v2 made them longer and loopable, and SFX v3 is aimed at full scenes with music and voices inside. Whether it delivers that will only show in direct tests against SFX v2, Firefly and Stable Audio 3.0 on the same prompts.

Sources

• SFX v3 announcement, ElevenLabs on X – release date, claims, availability in ElevenCreative and ElevenAPI
• Create sound effect, ElevenLabs API reference – v2 and v3 model IDs, default model, duration, loop, prompt influence, output formats
• Sound effects, ElevenLabs docs – 30-second limit, looping, 40 credits per second, prompting
• ElevenLabs changelog – October 19 sound effects endpoint update
• SFX v2 release, The Decoder – 30-second clips, looping, 48 kHz