One prompt, several sounds: ElevenLabs takes on the whole scene

Describe the scene, then check access

ElevenLabs says its new sound-effects generator follows several instructions in one prompt more reliably. Its launch demonstration moves from a cheering crowd to a ball dropping and then booing. The company offers it inside ElevenCreative and through its API, the interface developers use to connect it to other software.

The documentation puts a condition on that launch: access is limited and may vary by region. It does not name the eligible countries. An announcement of availability therefore does not establish access for a particular European account.

A longer clip can change the comparison

Tools publication AI for Mortals reports running 19 generations and publishes paired clips with their prompts. In one sequence, it asked for a door, gravel footsteps, a car door and an engine. With duration chosen automatically, v3 produced a 15-second clip; v2 chose two seconds and omitted parts of the sequence, according to the review.

When the reviewers gave v2 the same 15 seconds, they reported that it completed the sequence too. That result makes duration part of the explanation, rather than proving that a model change alone fixes every multi-sound prompt.

The reviewers used Gemini 2.5 Pro for sound annotations and adjusted playback levels. This is a small, AI-assisted comparison, not a blinded listener study or our own listening test.

The quiet cue may still need its own track

For a separate product-video example, the reviewers generated a click, a transition sound and a background layer individually, then placed them in an editor. They also reported that a small ball sound was weak beneath the crowd in both models’ outputs.

That distinction matters when a sound must land on a particular frame. Generating a scene can help sketch its atmosphere; arranging the individual cues remains a separate creative decision. It is another example of the work after generation that we followed in our campaign feature.