# ElevenLabs Voice Models Source: [ElevenLabs Voice Models]() Documentation index: [llms.txt]() Last updated: 2026-10-28 ChatCut’s inspected ElevenLabs route turns **text plus a selected voice into narration**. It defaults to `eleven_multilingual_v2`; official voices that advertise audio tags use `eleven_v3`. These are TTS calls, not voice conversion or the provider’s newer dialogue routes. Generate a project audio asset, then specify a timeline destination or add it manually. Exact spoken length is not guaranteed. **Last verified: 2026-10-28.** Evidence covers official documentation, ChatCut’s source contract at `b3cbb24` and separate bounded endpoint tests. No supplied test establishes a fresh production narration, timeline placement or audio-quality result. [中文版](https://pr-691-preview.chatcut.dev/zh/docs/models-elevenlabs). ## Models, controls and evidence | Model or control | Native capability and inspected ChatCut route | Evidence and source | | --- | --- | --- | | `eleven_multilingual_v2` | ChatCut’s inspected default TTS model. Native ceiling: 10,000 characters; 29 languages. | Official documentation and source review, 2026-10-28; [model overview](https://elevenlabs.io/docs/overview/models), [ChatCut voiceover workflow](https://pr-691-preview.chatcut.dev/docs/ai-voiceover). Exact TTS boundaries were not probed. | | `eleven_v3` | Use for official voices advertising audio tags. Native ceiling: 5,000 characters; 70+ languages. | Official documentation and source review, 2026-10-28; [model overview](https://elevenlabs.io/docs/overview/models), [TTS guidance](https://elevenlabs.io/docs/overview/capabilities/text-to-speech/best-practices). No matched narration test. | | `eleven_flash_v2_5` | Native TTS alternative with a 40,000-character ceiling and 32 languages. The adapter accepts a model ID; current selection and entitlement must be checked. | Native documentation, reviewed 2026-10-28; [model overview](https://elevenlabs.io/docs/overview/models). No live boundary or ChatCut production test. | | Delivery controls | Inspected ElevenLabs controls: speed 0.7–1.2 and stability 0–1. ChatCut does not expose native seed, pronunciation dictionaries or request-context fields here. | Source contract review, 2026-10-28; [TTS API](https://elevenlabs.io/docs/api-reference/text-to-speech/convert). These ranges belong to the ElevenLabs route. | | Output and length | The inspected adapter requests `mp3_44100_128`. It has no exact-duration field and does not automatically split oversized text. | Source review, 2026-10-28; [TTS API](https://elevenlabs.io/docs/api-reference/text-to-speech/convert), [ChatCut workflow](https://pr-691-preview.chatcut.dev/docs/ai-voiceover). Playback duration depends on the performance. | | V4 / V4 Turbo | Provider dialogue routes are not exposed by this ChatCut TTS adapter. Changing the model ID alone does not wire those endpoints. | Source and documentation review, 2026-10-28; [models](https://elevenlabs.io/docs/overview/models), [dialogue API](https://elevenlabs.io/docs/api-reference/text-to-dialogue/convert). ChatCut route access remains unverified. | | Voice changing | A short upstream `eleven_multilingual_sts_v2` conversion succeeded. The inspected ChatCut TTS tool has no source-audio input for it. | Bounded endpoint test and source review, 2026-10-28; [model identity](https://elevenlabs.io/docs/overview/models). This does not establish a ChatCut voice-changer workflow. | | Credits | Input-character usage and playback seconds are different quantities. Use the current estimate and completed ledger charge. | Reviewed 2026-10-28; [credit policy](https://pr-691-preview.chatcut.dev/docs/credits-policy), [generation credits](https://pr-691-preview.chatcut.dev/docs/generation-credits). No current retail cash tariff is quoted here. | ## Choose a voice before generating Choose a voice from **Voice Library** or **My Voices** and check its supported uses. A voice intended only for an avatar is not automatically suitable for standalone TTS. Selecting a voice attaches it to the draft; it does not generate narration until you submit text. Use the selected voice’s supported language and delivery guidance. For an official voice advertising audio tags, the inspected tool guidance specifies `eleven_v3`. Do not assume every model or voice supports the same tags, languages or controls. The native ceilings above are model-specific documented limits, not successful maximum-length tests on ChatCut. Voice cloning, TTS and voice changing are different operations. Cloning prepares a reusable voice from an authorized recording; TTS reads new text with a selected voice; voice changing transforms an existing performance. ChatCut’s inspected custom-voice path uses Fish Audio, so its cloning limits must not be presented as ElevenLabs limits. Follow [Voice Library / My Voices](https://pr-691-preview.chatcut.dev/docs/voice-library) for that separate workflow. ## From script to an editable audio clip 1 #### Prepare the script Finalize the words, choose the language and check the selected model’s character ceiling. Split long scripts into manageable sections before sending. 2 #### Select the voice Open + → Voices, preview the available voices and attach one that supports TTS. Follow its model and delivery guidance. 3 #### Set the destination Ask for narration on a named audio track at a timecode, or identify the clip it should accompany. Review the credit estimate and submit. 4 #### Audition and adjust Find the completed audio in My Assets. If it was not placed, drag it to the timeline. Check pronunciation, pauses and sync, then move or trim it. Example: “Read the following approved script with the attached voice. Keep the delivery calm. Add the completed narration to an available audio track starting at 00:03: Welcome to the next chapter.” The timecode is a placement instruction. It does not promise a particular generated length or sentence-by-sentence synchronization. The expected editable output is an audio asset and, when placed, a timeline clip. Selecting a voice alone does not spend generation usage; submitting narration uses credits. Verify a current quote rather than converting an upstream API price into ChatCut credits. Agent usage may be separate. ## Boundaries of the inspected adapter `submit_voice` sends a TTS request with text and one resolved voice. It does not become a speech-to-speech or dialogue tool when a different model ID is supplied. V4/V4 Turbo need the appropriate dialogue integration; those routes are **not exposed on ChatCut here**, while their actual entitlement remains unverified. The separate sound-effect route also has its own contract. ChatCut’s inspected tool and adapter cap requested SFX duration at **0.5–22 seconds**. Provider endpoint validation allowed **0.5–30 seconds**, but a full 30-second output was not generated. The 22-second ceiling is a ChatCut integration limit. The adapter omits the upstream model ID, so a job label does not prove which SFX model served it. See [music and sound effects](https://pr-691-preview.chatcut.dev/docs/music-and-sound-effects). ## Frequently asked questions Which voice and model should I choose? Choose a voice that supports TTS and the language you need. The inspected default is eleven\_multilingual\_v2; official voices advertising audio tags use eleven\_v3. Follow the selected voice’s guidance rather than assuming all provider models are offered. How much text fits? Documented native ceilings are 5,000 characters for v3, 10,000 for Multilingual v2 and 40,000 for Flash v2.5. These boundaries were not live-tested here. ChatCut does not automatically split an oversized TTS request. How do I place narration at a timecode? Include a destination such as an available audio track starting at 00:03. The completed asset appears in My Assets; if it was not placed, drag it to the timeline. Review timing afterward. How do usage and playback length differ? TTS usage is based on input characters, while playback duration varies with words, pauses and delivery. Check the current editor estimate and settled ledger; changing a timeline clip’s length is not a new TTS quote. ## Sources and related guides Official [models](https://elevenlabs.io/docs/overview/models), [TTS API](https://elevenlabs.io/docs/api-reference/text-to-speech/convert), [dialogue API](https://elevenlabs.io/docs/api-reference/text-to-dialogue/convert) and [sound API](https://elevenlabs.io/docs/api-reference/text-to-sound-effects/convert): snapshot accessed 2026-10-27; reviewed with the pinned ChatCut contract and bounded endpoint evidence on 2026-10-28. Provider capability does not certify ChatCut production availability. Continue with [AI voiceover for video](https://pr-691-preview.chatcut.dev/features/ai-voiceover), [voice isolation](https://pr-691-preview.chatcut.dev/features/ai-noise-removal), [model capability reference](https://pr-691-preview.chatcut.dev/docs/model-capability-matrix), [generation references](https://pr-691-preview.chatcut.dev/docs/generation-references) or [workflow failure recovery](https://pr-691-preview.chatcut.dev/docs/workflow-failures).