ElevenLabs has released SFX v3, the new version of its sound effects model. From a short text description, it generates a sound clip of up to 30 seconds. ElevenLabs offers SFX v3 in ElevenCreative, the company's creative platform, as well as in the mobile app and via the ElevenAPI programming interface. In the launch blog post, ElevenLabs focuses above all on prompts that describe several sounds at once.
What can SFX v3 do better than SFX v2?
A soccer clip needs cheering spectators, a bouncing ball, and, seconds later, boos from the same crowd. Such mixes previously took several attempts, ElevenLabs writes. SFX v3 follows multiple instructions within a single prompt more reliably, so a larger share of the described sounds actually ends up in the clip.
What SFX v3 can and cannot do
New sound effects model from ElevenLabs, presented on October 29, 2026
- Implement multiple sounds in one prompt more reliably
- Combine music and vocals with everyday noises
- Gongs, chimes, and whistles with fewer artifacts
- Effects longer than 30 seconds per run
Improvements according to ElevenLabs benchmarks, independent tests are lacking
Sources: ElevenLabs Blog, ElevenLabs Documentation
In addition, the model produces music and vocals more convincingly, for example a short melody together with everyday noises. With gongs, chimes, and whistles, there are fewer artifacts that make a tone sound shrill or crackly. The improvements have so far only been substantiated by ElevenLabs' own benchmarks; independent tests are lacking.
With SFX v2, ElevenLabs introduced clips of up to 30 seconds, seamless loops, and a sampling rate of 48 kHz in September 2025, as The Decoder reported at the time. v3 changes nothing about these key data: the API documentation still lists a maximum of 30 seconds per run and WAV files at 48 kHz. The new version thus improves accuracy, not length.
Where do you use SFX v3, and what does an effect cost?
In ElevenCreative, you generate effects in the Sound Effects tab, in Flows, in Studio, and in the Music Editor. Via the API, you select the model with the identifier eleven_text_to_sound_v3. However, the documentation adds a caveat: SFX v3 is only available to a limited extent and may not be available in every region. ElevenLabs does not reveal which regions are missing, not even for Germany.
Billing is in credits. The pricing page sets 200 credits per generated effect. If you specify a fixed duration, each second costs 40 credits, so a 30-second effect costs 1,200 credits. ElevenLabs does not state a separate price for v3. The free plan includes 10,000 credits per month, which is enough for 50 effects; the Creator plan with 121,000 credits is enough for 605 effects.
You may only use the effects commercially with a paid plan from Starter upward; the free plan is limited to private use. ElevenLabs does not list euro prices even on the German pricing page.
What this means for you
For videos with a dense soundscape, a comparison is worthwhile. It's best to take a prompt that SFX v2 has failed at so far and listen to how many of the described sounds v3 now implements. ElevenLabs advises starting with the most important sound, adding the ambient sounds afterward, and adjusting the description until the result matches your vision.
Before you build an effect into a client video, quickly check whether your account is on at least the Starter plan. If you already work with Adobe, Firefly is an alternative, since the program also generates sound effects with AI; how that works is shown in the lesson on sound effects with Adobe Firefly.
Whether v3 is already enabled for your account, you can clarify via the API with a single request: with the identifier eleven_text_to_sound_v3, a request only runs if access is enabled. The AI Guide for Video, Audio and Voice presents further AI tools for sound and voice.
ElevenLabs presented SFX v3 on October 29, 2026.
Sources
