ElevenLabs launches SFX v3, a sound effects model for multi-sound scenes in ElevenCreative and ElevenAPI

ElevenLabs has released SFX v3, the third generation of its text-to-sound-effects model, and made it available both in ElevenCreative, the company's suite of creative tools, and through ElevenAPI for developers. The announcement, published on the ElevenLabs blog on October 29, 2026, presents the model as the company's newest sound effects system and focuses on a specific weakness of earlier generators: scenes that need several distinct sounds in a single clip rather than one isolated effect.

The company frames the problem with a sports example. A short football clip may need a crowd cheering, a ball dropping and, a few seconds later, the same crowd booing. Earlier text-to-audio systems often handled the first or the most salient element of such a description and dropped the rest, which forced editors to generate each layer separately and mix them by hand. ElevenLabs says SFX v3 follows several instructions in one prompt more reliably, so that more of the described sounds actually appear in the output.

Alongside multi-sound prompts, the release claims two further improvements over SFX v2: more convincing generation of musical phrases and vocal performances, including when they are combined with environmental sounds, and cleaner, more natural audio with fewer unwanted artifacts on sounds such as gongs, chimes and whistles. ElevenLabs states that SFX v3 outperforms SFX v2 in benchmark testing, but the announcement does not publish the benchmark, the metrics used or any scores, so that comparison remains a vendor claim that cannot be checked independently at this stage.

··········

WHAT SFX V3 CHANGES COMPARED WITH SFX V2.

The central change is prompt adherence for compound scenes. In the company's description, the model is better at parsing a prompt that contains several events and rendering each of them, which matters for video editors, game developers and podcast producers who typically describe a moment rather than a single sound. ElevenLabs recommends a simple workflow for writing such prompts: start with the main action, then add the surrounding sounds, and refine the description from there. The approach mirrors how a sound designer builds a scene, with a primary event in the foreground and ambience or secondary reactions layered around it.

The second area is music and voice. Sound effects models have historically struggled when a prompt asks for something tonal, such as a short musical sting, a choir note or a human vocal reaction, because these sounds expose pitch errors and synthetic timbre more clearly than noise-like effects such as wind or footsteps. ElevenLabs says SFX v3 produces musical phrases and vocal performances more convincingly and can pair them with environmental audio, for example a vocal performance placed inside an ambient scene. The company does not specify whether this extends to intelligible lyrics or speech, and nothing in the announcement suggests that SFX v3 is meant to replace the company's dedicated music or text-to-speech models.

The third claim concerns audio quality. Metallic and resonant sources such as gongs, chimes and whistles are a known stress test for generative audio, because artifacts in the decay of the sound are easy to hear. ElevenLabs says SFX v3 is less likely to introduce such artifacts and that the cleaner output helps effects sit naturally in a mix, whether they are placed in the foreground or used as background texture.

........

Area	SFX v2	SFX v3 (as described by ElevenLabs)
API model ID	eleven_text_to_sound_v2	eleven_text_to_sound_v3
Default in the API	Yes	No, must be selected
Multi-sound prompts	Baseline	More of the described sounds appear
Music and vocals	Baseline	More convincing phrases and vocal performances
Artifacts on gongs, chimes, whistles	Baseline	Less likely, per the company
Published benchmark scores	Not applicable	Not disclosed

........

··········

WHERE SFX V3 IS AVAILABLE AND WHO CAN USE IT.

For creators, the model is exposed in several parts of ElevenCreative. According to the announcement, sound effects can be generated from text prompts in Flows, in the dedicated Sound Effects tab, in Studio and in the Music Editor, and the company also points users to its mobile app. This means that someone editing a narrated video in Studio or assembling a soundtrack in the Music Editor can generate effects in place rather than switching to a separate tool and importing the files afterwards.

For developers, SFX v3 is part of ElevenAPI. The developer documentation, however, adds an important qualification that the blog post does not mention: it describes SFX v3 as being in limited access and says it may not be available in every region. Accounts that do not yet have access receive an "Unknown model" error when they request it. Teams planning to build on the new model should therefore test the model ID against their own account before committing a production feature to it, and keep SFX v2 as a fallback where availability is uncertain.

The documentation does not list which subscription tiers receive SFX v3 first, and the announcement does not mention a waitlist or a regional rollout schedule. ElevenLabs' public sound effects page continues to list commercial use rights on paid plans (Starter, Creator and Pro) and personal use only on the free plan, a distinction that applies to anyone who intends to use generated effects in client work, monetized videos or commercial games.

........

Surface	SFX v3 status
ElevenCreative Sound Effects tab	Available, per the announcement
ElevenCreative Flows	Available, per the announcement
ElevenCreative Studio	Available, per the announcement
Music Editor	Available, per the announcement
ElevenLabs mobile app	Mentioned as a way to try it
ElevenAPI	Limited access, may not be available in every region (developer docs)

........

··········

HOW DEVELOPERS CALL SFX V3 IN ELEVENAPI.

SFX v3 uses the existing sound generation endpoint rather than a new API. Requests go to POST https://api.elevenlabs.io/v1/sound-generation, and the model is selected through the model_id field of the request body. The API reference lists two allowed values, eleven_text_to_sound_v2 and eleven_text_to_sound_v3, and the default remains v2. Existing integrations will therefore keep producing SFX v2 output until developers change the parameter explicitly, which avoids silent changes in the character of the generated audio for products already in production.

The other request parameters are unchanged. The text field carries the prompt and is the only required field. The duration_seconds parameter accepts values up to 30 seconds; the API reference gives a minimum of 0.5 seconds, while the sound effects capability page cites 0.1 seconds, and when the field is left empty the model infers a duration from the prompt. The loop flag, false by default, asks for an effect designed for seamless repeated playback, which is the documented route for ambiences longer than the 30-second cap. The prompt_influence parameter, between 0 and 1 with a default of 0.3, controls how literally the model follows the prompt: higher values reduce variability, while lower values allow more creative variations.

The following request is an illustrative example built only from the documented endpoint and parameters; it is not copied from ElevenLabs' documentation.

curl -X POST "https://api.elevenlabs.io/v1/sound-generation" \
  -H "xi-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Stadium crowd cheering, a football bounces twice on the turf, then the crowd boos", "model_id": "eleven_text_to_sound_v3", "duration_seconds": 8, "prompt_influence": 0.5}'

Output formats are set through the output_format query parameter, with mp3_44100_128 as the default. The capability page states that all effects can be exported as MP3 and that non-looping effects are also available as 48 kHz WAV, while the API reference notes that some formats are tied to plan levels: 192 kbps MP3 requires the Creator tier or above, and PCM and WAV output at 44.1 kHz require the Pro tier or above. The response returns the encoded audio together with a character-cost header used for billing.

........

Parameter	Type	Default	Documented range or values
text	string	Required	The sound effect prompt
model_id	enum	eleven_text_to_sound_v2	eleven_text_to_sound_v2, eleven_text_to_sound_v3
duration_seconds	number or null	null (inferred)	Up to 30 seconds
loop	boolean	false	Seamless looping output
prompt_influence	number or null	0.3	0 to 1
output_format	query enum	mp3_44100_128	Some formats require Creator or Pro

........

··········

WHAT SOUND EFFECTS COST ON ELEVENLABS.

ElevenLabs has not announced a separate price for SFX v3. Its API pricing page lists sound effects at $0.12 per minute of generated audio, the same rate on every plan, without distinguishing between model versions. At that rate a 5-second effect costs about $0.01, a 10-second effect about $0.02 and a maximum-length 30-second effect about $0.06. The sound effects capability page expresses the same cost in credits for requests with an explicit duration, at 40 credits per second, which puts a 10-second clip at 400 credits and a 30-second clip at 1,200 credits.

Translated into the monthly credit allowances shown on the ElevenLabs sound effects page, the free plan's 10,000 credits correspond to about 250 seconds of generated effects, or roughly 25 ten-second clips. The Starter plan, at $6 per month, includes 30,000 credits, equal to about 75 ten-second clips. Creator, at $22 per month, includes 121,000 credits, or about 302 ten-second clips, and Pro, at $99 per month, includes 600,000 credits, or about 1,500 ten-second clips. These figures assume that all credits are spent on sound effects; in practice, the same pool is shared with text to speech, dubbing and the other ElevenLabs tools on each plan.

The cost model has a practical consequence for iterative work. Because each attempt is billed by length, a sound designer who generates four variations of an 8-second scene spends the same as one 32-second generation. The claimed gain in multi-sound adherence, if it holds in practice, could reduce the number of attempts needed to get a usable layered effect, which is where the most immediate savings for heavy users would come from.

........

Clip length	API cost at $0.12 per minute	Credits at 40 per second
5 seconds	about $0.01	200
10 seconds	about $0.02	400
30 seconds (maximum)	about $0.06	1,200

........

··········

WHERE SFX V3 FITS IN THE ELEVENLABS PRODUCT LINE.

Sound effects have become a natural complement to the voice and music products that ElevenLabs sells to video creators, game studios and app developers. A narrated explainer, a dubbed trailer or a voice agent demo often needs background ambience, transitions and short stings, and producing them inside the same platform shortens the path from script to finished audio. Integrating SFX v3 into Studio and the Music Editor reflects that logic: the model is presented less as a standalone generator and more as a component of complete audio projects.

For game developers and interactive applications, the API route matters more. Procedural or on-demand generation of effects, for instance ambient loops for different environments or variations of an impact sound, depends on predictable behaviour, clear limits and stable pricing. The documented 30-second cap, the looping option and the per-second billing are the parameters that determine whether such use is viable, and SFX v3 changes none of them. What it changes, according to ElevenLabs, is how faithfully the model renders a described scene.

··········

WHAT IS ESTABLISHED AND WHAT REMAINS UNCERTAIN.

The established facts are concrete. SFX v3 was announced on October 29, 2026, it is available in the ElevenCreative surfaces named by the company, it is reachable in the API through the existing sound generation endpoint with the model ID eleven_text_to_sound_v3, and SFX v2 remains the API default. The documented limits are unchanged, with clips of up to 30 seconds, an optional looping mode and the same output formats, and sound effects are priced at $0.12 per minute of audio on the API pricing page.

Several points are still open. ElevenLabs states that SFX v3 beats SFX v2 in benchmark testing but has not published the evaluation, so the size of the improvement in multi-sound adherence, musical content and artifact reduction cannot yet be quantified. The developer documentation describes API access as limited and possibly unavailable in some regions, which sits uneasily with the blog post's message of general availability, and neither source says which plans or regions are included. Until ElevenLabs clarifies the rollout and publishes evaluation details, developers should treat SFX v3 as an opt-in upgrade to test against their own prompts, while creators working inside ElevenCreative can compare it directly with earlier output on the scenes that previously required manual layering.

··········

FOLLOW US FOR MORE.

·····

DATA STUDIOS

·····

datastudios.org