ElevenLabs releases SFX v3 sound-effects model
ElevenLabs has added SFX v3 to its Sound Effects API, which follows multi-part prompts more closely and clips less on high-frequency sounds, though the company describes it as an incremental update.
ElevenLabs announced SFX v3, the newest model in its Sound Effects API. The company says it follows prompts that contain several instructions more closely than SFX v2 and handles vocals and music better.
Developers can call the model by setting model_id to eleven_text_to_sound_v3 on the existing POST /v1/sound-generation endpoint. The model chooses how long each clip runs based on the prompt. Callers who need a set length can pass duration_seconds. The prompt_influence parameter defaults to 0.3, and ElevenLabs suggests raising it when outputs drift from the request: higher values track the prompt more closely and vary less between generations. The output_format parameter sets the codec and sample rate, for example mp3_44100_128 or pcm_48000.
According to ElevenLabs, v3 is less likely to clip on high-frequency sounds such as gongs, chimes and kettle whistles, and its engineering team says the output sounds more natural overall. The post does not include benchmarks or quantitative comparisons.
There is one limitation. Looping (loop: true) still works only with eleven_text_to_sound_v2, so applications that need looped audio must stay on v2.
For prompts with several parts, ElevenLabs advises listing events in the order they should happen, such as "Footsteps on gravel, then a metallic door opens." For a single sound, it recommends a short, concrete prompt. It also notes that wording affects results: one tester got better output from "arguing" than from "fighting."
Both models run on the same endpoint, so ElevenLabs suggests running common prompts through v2 and v3 before switching. SFX v3 is also available in the Sound Effects tab, Flows and Studio in ElevenCreative.