ElevenLabs released SFX v3, a new version of its sound effects model, on October 29, 2026. The company says it follows prompts that describe several sounds more reliably, does better with music and vocals, and makes cleaner audio than SFX v2. It is live in ElevenCreative, the company's app for creators, and in ElevenAPI for developers. Here is what changed, how to use it, and what it costs.
What ElevenLabs says is new
The launch post, written by Renee Zhang and Eric Michaelis, names three improvements.
- Several sounds in one prompt. ElevenLabs says SFX v3 is "better at following several instructions within a single prompt," so more of the sounds you describe make it into the clip. Its example is a football scene with a crowd cheering, a ball dropping, and the same crowd booing seconds later. The company says that kind of scene used to take trial and error.
- Music and vocals. The model "generates music and vocals more convincingly," per the post. ElevenLabs pitches this for effects that pair a musical phrase or a vocal performance with everyday or environmental sounds.
- Fewer artifacts. ElevenLabs says SFX v3 is less likely to produce unwanted artifacts on sounds such as gongs, chimes and whistles, so a gong keeps its resonance and a whistle holds its tone "without becoming harsh or crackly."
The post says SFX v3 produces cleaner, more natural audio than SFX v2 "in benchmark testing." It does not name the benchmark or publish scores, so treat the quality claims as the company's own until independent tests appear.
Where you can use it
In ElevenCreative, the post lists four places to generate effects from a text prompt: Flows, the Sound Effects tab, Studio, and the Music Editor. SFX v3 is also in the ElevenLabs mobile app.
For developers, the Create sound effect API reference lists two allowed values for model_id: eleven_text_to_sound_v2 and eleven_text_to_sound_v3. The default is still eleven_text_to_sound_v2. If you call the endpoint without a model ID, you get the older model, so set model_id to eleven_text_to_sound_v3 to use the new one.
The settings that matter
The sound effects documentation and the API reference describe the controls for the sound effects endpoint. These pages describe the endpoint in general, not SFX v3 alone.
- Duration. The capability docs give a range of 0.1 to 30 seconds; the API reference says duration_seconds must be at least 0.5 and at most 30. If you leave it out, the model picks a length based on the prompt. The docs say a set duration costs 40 credits per second.
- Looping. A
loopoption makes a clip that plays on repeat without a clear start or end point. The docs suggest it for ambience, such as 30 seconds of soft rain looped under an audiobook or game scene. - Prompt influence. A value from 0 to 1, with a default of 0.3. Higher values follow the prompt more literally and give less variety between takes. Lower values allow more creative variation.
- Output. MP3 for all effects, and WAV at 48 kHz for non-looping effects, per the docs. The API's changelog adds a list of WAV sample rates from 8 kHz to 48 kHz.
How to write a multi-sound prompt
The multi-sound claim is the headline, so the way you write prompts matters. The ElevenLabs prompting guide gives three patterns.
- Simple effects: short and concrete, such as "Glass shattering on concrete" or "Heavy wooden door creaking open."
- Sequences: describe the order of events, such as "Footsteps on gravel, then a metallic door opens" or "Sword being drawn, then clashing with another blade."
- Musical elements: name the style and the tempo or key, such as "90s hip-hop drum loop, 90 BPM" or "Vintage brass stabs in F minor."
The guide also lists audio terms the model understands, including impact, whoosh, ambience, one-shot, loop, stem, braam, glitch and drone. The launch post suggests a workflow for scenes: start with the main action, add the sounds around it, listen, and then adjust the description.
One limit to keep in mind: the docs say the endpoint can make drum loops, bass lines and melodic samples, but for full music production they point to the separate Music API. SFX v3 is for effects and short musical moments, not whole songs.
What it costs
The ElevenLabs API pricing page lists Sound Effects at $0.12 per minute (as of October 29, 2026), with royalty-free output as MP3 or WAV. The plan table on that page lists included generations from 8 at the lowest tier to 30,000 at the highest. Its FAQ says Sound Effects are billed per generation. The pricing page does not list a separate price for SFX v3, and ElevenLabs did not mention a price change in the launch post.
Early reactions
Two posts on X in the first hours showed people testing layered prompts. User @GesoraMeshack wrote that SFX v3 "handled 'door opens, footsteps on gravel, car door shuts, engine starts' in one go. Four things in, four things back." User @kimiatehrani posted a night beach scene with waves, a campfire, a guitar and distant laughter, and said the model "kept the layers separate enough that I could hear each one." These are single user tests, not controlled comparisons.
Why it matters
ElevenLabs itself says layered scenes used to take trial and error. If SFX v3 holds up on multi-sound prompts, creators can get a full scene from one description instead of stitching several clips together. It also fits a pattern at ElevenLabs of putting voice, music and effects in one place: the company's Instant Soundtracks feature already generates music and timed sound effects for uploaded video, and ElevenLabs Cue lines up text with audio for captions and lyrics. For a wider look at what these tools cost, see our AI voice and music pricing comparison and our guide to the best AI voice generators.
Developers who want the new model should update model_id in their calls, since the API default has not moved to v3 yet.
