In this article
ElevenLabs announced SFX v3 on October 29, 2026, with an emphasis on generating several described sounds in one prompt. The company also reports improvements to musical and vocal elements and to the quality of effects such as chimes and whistles.
The release is relevant to video creators because a believable scene usually contains more than its loudest action. Background ambience, a nearby object and a small transition can each carry part of the story.
What the announcement changes
SFX v3 is offered through ElevenCreative and ElevenAPI. ElevenLabs’ documentation notes that access is limited and may vary by region. Check availability in the service you use before building a production plan around the model.
The reported quality improvements are ElevenLabs’ claims, not results from Tong benchmark tests. A useful evaluation is whether the sounds requested in your own brief are present, distinct and appropriate for the picture.
Write a sequence, not a list of adjectives
An original brief might read: “A wooden door closes softly, a small bell rings once nearby, and steady rain continues outside.” That gives a sound designer three roles: an action, a short foreground accent and an ongoing background.
Listen for the relationship between those layers. Does the bell compete with speech? Does the rain overwhelm the room? If a specific cue needs exact timing, keep that timing decision in the edit and check the generated effect against the picture.
Keep the sound tied to the story
Save the prompt and the selected audio version so later revisions have a clear reference. Compare candidates at a similar listening level; a louder file can seem more exciting without being a better fit.
For the visual side of a sequence, our cinematic AI video guide helps connect action and camera direction. Tong’s model directory lists its own supported creation tools; the SFX v3 availability described in this update is in ElevenLabs.