ElevenLabs SFX v3 generates multi-sound scenes from one prompt

ElevenLabs released SFX v3 on October 29, adding a sound-effects model designed to follow several sounds in a single prompt. The company says it improves multi-part instructions, music and vocals, and audio quality compared with SFX v2. It is available through ElevenCreative and ElevenAPI.

For video and game production, the change allows a whole sound scene to be requested together, or the new model to be selected in an existing integration. Availability has conditions: the official model catalog labels SFX v3 limited access and says it may be unavailable in some regions. Accounts without access receive an Unknown model error. SFX v2 remains the API default.

One company example combines a cheering football crowd, a ball dropping and the crowd booing. ElevenLabs also reports less clipping on high-frequency sounds such as chimes and whistles. These are vendor claims; the published account provides no numerical improvement or independent evaluator score for the comparison.

The web product guide describes four variations per generation, with results retained in History and an option to save favorites. The text field accepts up to 450 characters. The web guide lists MP3 at 44.1kHz and WAV at 48kHz, with regeneration available after revising the description or settings.

Up to 30 seconds, with looping

The API specification accepts a duration from 0.5 to 30 seconds. Leaving it unset lets the model choose a length from the prompt. Prompt influence ranges from zero to one and defaults to 0.3; higher values follow the description more closely while reducing variation. SFX v3 uses the existing sound-generation endpoint with a different model selection.

SFX v3 supports seamless looping. The capability guide describes repeating a short rain ambience clip for a longer background track, while retaining a maximum of 30 seconds per generated clip. Musical elements can include drum loops and short melodic samples; full-song production has a separate Music API. The guide cautions that instructions requesting an exact number of repetitions are not always followed.

The current monthly price list includes 10,000 credits on Free, 30,000 for $6 on Starter and 121,000 for $22 on Creator. Those credits are shared across products, including speech and music. Its usage table lists sound effects at approximately 200 credits per generation. Listed prices exclude taxes.

API costs depend on request settings. The capability guide specifies 40 credits per second when a duration is supplied, while the product guide says charges vary between the website and API with the number and length of outputs. Under that stated rate, a ten-second API generation calculates to 400 credits. The endpoint defaults to 44.1kHz, 128kbps MP3. Its specification requires Creator or higher for 192kbps MP3, and Pro or higher for 44.1kHz PCM and WAV.

Commercial use requires a paid account. The sound-effects product page permits royalty-free use in projects such as advertising, YouTube videos and social content for paid users, while describing Free as noncommercial. The web guide offers MP3 and WAV downloads. Those usage permissions and the endpoint’s format-specific subscription requirements apply separately.

Unused paid credits can roll over for up to two months, taking the balance to at most three monthly allowances including the current one. Downgrading or cancelling forfeits unused paid credits at the end of the billing cycle; Free has no rollover. Separately purchased pay-as-you-go credits fall outside that rollover cap. Credits are charged on generation, rather than download. Some regenerations may be free, as indicated before the request.