ElevenLabs announces SFX v3, the latest version of its sound effects generation model. Available now in ElevenCreative and via ElevenAPI, it is presented as more reliable for prompts describing multiple sounds at once, with more convincing music and voices, and cleaner, more natural sound than SFX v2 in benchmark tests.
A powerful scene rarely relies on a single sound. A football sequence may require a cheering crowd, a ball bouncing, then the same crowd booing a few seconds later. Until now, obtaining all these elements from a single prompt required multiple attempts.
Multiple sounds in a single prompt
SFX v3 better follows multiple instructions contained within the same text. Therefore, more of the described sounds are present in the generated audio. According to ElevenLabs, the user can describe the sounds that make up their scene and listen to how they interact with each other.
For creators of videos, games, or other projects, the proposed method involves starting from the main action, adding the surrounding sounds, then listening to their combined output. The description can then be adjusted to get closer to the desired result.
Music, voices, and sound quality
The model also improves generation when the prompt includes music and voices. It allows, according to the company, for creating effects that combine a musical phrase or vocal performance with ambient and everyday sounds.
The quality of each sound influences a scene’s credibility. A gong must retain its resonance, a bell its ring, and a whistle must hold its note without becoming shrill or crackly. ElevenLabs indicates that SFX v3 improves overall sound quality and reduces the likelihood of these unwanted artifacts on gongs, bells, and whistles. The resulting audio better preserves the character of the requested sound, whether it occupies the foreground or adds details in the background.
Availability
In ElevenCreative, effects can be generated and explored in the Sound Effects tab, added to Flows, or integrated into projects in Studio. SFX v3 is also accessible from the ElevenLabs mobile app. Developers can use the model via ElevenAPI, its programming interface, to integrate sound effects generation into their own products and workflows.
