ElevenLabs Launches SFX v3, a Sharper Text-to-Sound-Effects Model

ElevenLabs has released SFX v3, the newest version of its text-to-sound-effects model, and it is live now in its ElevenCreative app and through its developer API.

I'm covering this one because a lot of the people reading this blog make things: fan edits, AMVs, small indie games, visual novels. Sound is usually the part that gets left until the end, and a tool that can sketch a whole scene from a sentence is worth knowing about, even if you end up replacing it later.

What changed

According to ElevenLabs, SFX v3 does three things better than SFX v2:

• Several sounds in one prompt. It follows prompts that describe a sequence, so a line like a crowd cheering, a ball dropping and the crowd booing should come back as one clip instead of three you stitch together.
• Music and vocals. Prompts that mix a musical phrase or a voice with everyday sounds are handled more convincingly.
• Cleaner high sounds. The company says chimes, gongs and whistles are less likely to come out harsh or crackly.

ElevenLabs says the new model beats v2 in benchmark testing, but it hasn't published the scores, so I'd treat that as a claim until people compare them on their own prompts.

How to try it

In ElevenCreative it shows up in the Sound Effects tab, Flows, Studio and the Music Editor, and in the mobile app. Developers pick it with the model id eleven_text_to_sound_v3; the older v2 is still the default if you don't set one. Clips run up to 30 seconds, can loop seamlessly for ambience, and export as MP3 or 48 kHz WAV. The documentation notes that v3 is in limited access and may not be available in every region yet.

For anyone building a small game or a short animated piece, the looping ambience alone could save an evening of digging through sound libraries.

Source: ElevenLabs blog