ElevenLabs has launched SFX v3, the newest version of its text-to-sound-effects model, built for prompts that describe several sounds at once. The model follows multi-part instructions more reliably than its predecessor, generates music and vocals more convincingly, and produces cleaner, more natural audio in benchmark testing. It is available now in ElevenCreative and through ElevenAPI, as well as in the company’s mobile app.
## One Prompt, a Whole Scene
Big scenes rarely come down to one sound. A football clip might need a crowd cheering, a ball dropping, and the same crowd booing seconds later. Until now, getting all of that from one prompt took trial and error and patience.
Sounds would drop out or arrive in the wrong order on the first try, and fixing the result meant generating each effect separately. SFX v3 is built for that problem, keeping more of the sounds you describe in the finished audio.
The improvement is practical rather than cosmetic. Creators working on videos, games and other projects can now describe a scene as a whole, starting with the main action and adding the surrounding sounds around it. They can then adjust the description and listen to how the elements work together, instead of generating each effect separately and stitching them by hand.
## Music, Vocals and Cleaner Sound
SFX v3 also improves generation for prompts that include music and vocals. That opens up effects pairing a musical phrase or a vocal performance with environmental and everyday sounds, a combination the previous version handled less convincingly. The change matters for creators who want a scene to feel scored rather than assembled from separate clips.
The quality of an individual sound can change how believable a scene feels. A gong needs its resonance, a chime needs its ring, and a whistle should hold its tone without turning harsh or crackly. Getting that detail right is what separates a believable effect from a distracting one.
SFX v3 is less likely to produce those unwanted artefacts, so the result better preserves the character of the sound you asked for. ElevenLabs says the model outperforms SFX v2 in benchmark testing, though it has published no scores to back that claim yet.
## Where You Can Use It
SFX v3 is available now in ElevenCreative, where creators can generate effects from a text prompt in Flows, the Sound Effects tab, Studio and the Music Editor. Developers can reach the same model through ElevenAPI, bringing sound-effects generation into their own products and workflows. The model is also available on the go in the ElevenLabs mobile app.
## Our Read
SFX v3 is a quieter launch than a voice or music model, but it solves a bottleneck: sound design has been the fiddliest part of AI-generated video, because a scene is a stack of sounds. Keeping a cheering crowd, a bouncing ball and a booing crowd in sequence is the difference between a usable take and an afternoon of editing.
The timing also fits a pattern. ElevenLabs has shipped a rapid run of creator tools this autumn, from [Instant Soundtracks](https://geroundtable.com/elevenlabs-instant-soundtracks-video-music/) for video music to [Eleven Realtime](https://geroundtable.com/elevenlabs-eleven-realtime-speech-to-speech-models/) for voice agents. SFX v3 fills the gap between those two in sound design.
It covers the small, scene-level sounds that make generated video feel finished. The risk is that benchmark claims without numbers are hard to verify, so the model’s real test will be what creators hear in practice in the coming weeks.
## Why It Matters
Sound is where AI-generated video gives itself away. Flat or mismatched audio breaks immersion faster than a flawed frame. A model that handles layered prompts in one pass lowers the cost of sound, which matters most for small teams who cannot hire a sound designer. It also strengthens ElevenLabs’ position as a full-stack audio company rather than a voice specialist.
## Background
ElevenLabs launched SFX v2 in September 2025, offering 48kHz sound effects from text prompts with 30-second generation and royalty-free use. The [SoundCloud partnership](https://geroundtable.com/elevenlabs-soundcloud-ai-music-publishing/) announced on 5 November lets creators publish AI-generated music directly to the streaming platform. SFX v3 extends the same creator-first strategy into the finer grain of sound design.
Source: [ElevenLabs blog](https://elevenlabs.io/blog/sound-effects-v3)
0 Comments