ElevenLabs SFX v3 builds richer sound from one prompt
The new sound-effects model is in ElevenCreative and the API, with stronger multi-sound prompts, music, and vocals. The post does not list a price.
Every ElevenLabs update in one feed — release notes, features and changes, straight from the source.
104 updates
The new sound-effects model is in ElevenCreative and the API, with stronger multi-sound prompts, music, and vocals. The post does not list a price.
ElevenAgents can switch supported LLMs to a fast tier for lower latency at a higher per-token price, and the usage calculator now estimates the cost.
Website crawls for ElevenAgents knowledge bases accept up to 50 exclude patterns, set through the API or the whole-website crawl dialog.
The new 0 to 1 setting for v4 Professional Voice Clones replaces the deprecated use_pvc_as_ivc flag on Text to Dialogue requests.
Service account keys in multi-seat ElevenLabs workspaces can expire after 15 minutes to 90 days, set by API field or dashboard.
Eleven Realtime Translation lets two people each speak their own language on a call, in 90+ languages. The alpha is limited to enterprise customers.
Eleven Realtime v1 Mini and v1 Max each work as a complete voice agent, starting to talk in about half a second in 90+ languages.
Two models, Mini and Max, bundle the LLM, speech-to-text and text-to-speech settings for ElevenAgents, with an OpenAI Realtime-compatible API.
Enterprise workspaces can stream speech through a WebSocket API and get translated speech back in a chosen voice, with optional Zero Retention Mode.
The new API scores up to 10 videos (200 MB combined) as one, with optional sound effects and separate music and SFX tracks.
Workspace admins can tune error-rate thresholds per agent and route alerts to webhooks, Slack or PagerDuty, with default monitors running automatically.
Enterprise customers can set enable_logging=false on music and sound effects API calls, and Sound Effects can now return WAV audio.
Speech Engine conversations can now run to 40 minutes, up from the 10-minute default, and the list endpoint can sort by last update.
The dubbing language endpoint now returns a WebVTT subtitle URL for each dubbed language, signed and valid for one hour.
A new declined_resubmittable value in voice sharing's review_status marks declined voices that can be sent back for another review.
Workspace admins can switch off a service account key through the API, including after a plan downgrade, and group listings now show concurrency limits.
The @elevenlabs/client 1.28.0 release adds an experimental persistentSession option and an onConversationHistory callback for resuming WebSocket text chats.
Merging moves conversations onto the target ticket and raises its priority to the highest among the merged tickets.
The listing endpoint returns collections the caller can access, sorted by title and including permission details.
Workspace administrators must enable the fallback, which defaults to false and requires an available basic seat.
Python gains iterable audio playback, while Scribe adds document context and conversation widgets gain carousels and end confirmation.
The Alpha analyzes agent conversations, proposes versioned changes and creates simulations, with approval required before deployment.
Integrations can drop separate compile calls, while invalid procedures return per-procedure errors.
Python gets the same agent clients, while realtime packages add transcript editing and Swift preserves JSON types in dynamic variables.
The two models cover 90+ languages, and cloned voices now speak other languages without carrying over their original accent.
Agent builders can now pick from six new LLMs, and test runs can report credits and USD cost.
The API lists published Flows templates, starts runs on a chosen version and can post each run's final result to a webhook.
Text to Dialogue API requests can now carry context and request IDs between generations, and use the Instant Voice Clone version of a Professional Voice Clone.
The JavaScript and Python SDKs both reach v2.70.0 with transcript editing types, and the CLI fixes an API key and OAuth credential conflict.
OpenAI's Sora API retires on 24 September and ByteDance drops Seedance 1.5 Pro on 11 November, so affected flows need a new video model.