AI Voiceover
Turn a script and selected voice into narration, then place, review and trim it on your video timeline.
ChatCut turns a script and selected voice into a project audio asset. Choose a voice, supply the words and say where narration should start; if it is not placed automatically, drag it from My Assets to an audio track. The inspected ElevenLabs route is text-to-speech. Character limits, delivery controls and language coverage depend on the selected model and voice.
Last verified: 2026-10-28. This is a documentation and source-contract review, not a new production narration or placement test. Exact ElevenLabs identities and evidence are in the model reference.
Facts to check before generating
| Question | Inspected route or documented behavior | Source |
|---|---|---|
| Which model? | The inspected ElevenLabs default is eleven_multilingual_v2; audio-tag official voices use eleven_v3. | 2026-10-28 source review; model reference, ElevenLabs models. |
| How much text? | Native ceilings: Multilingual v2 10,000 characters, v3 5,000. ChatCut does not automatically split oversized requests; exact boundary generation was not tested. | 2026-10-28 review; model reference, official models. |
| Which controls? | Inspected ElevenLabs speed 0.7–1.2 and stability 0–1. Other engines have separate controls; do not apply a universal speed range. | 2026-10-28 source review; model reference, TTS API. |
| Where does it go? | My Assets; placement depends on the requested destination. Selecting a voice alone does not generate audio. | ChatCut voiceover guide, reviewed 2026-10-28. |
| How is it charged? | Input-character usage differs from playback seconds. Check the current estimate and settled ledger, rather than a universal price per minute. | Credits Policy, Generation Credits, reviewed 2026-10-28. |
Which voice and model should I choose?
Use + → Voices, preview Featured or My Voices, then attach a voice that supports standalone TTS and the language you need. Follow the voice’s model and delivery guidance. The voice library guide also explains the separate authorized voice-cloning workflow.
TTS reads new text. Voice changing transforms a source performance, and cloning prepares a reusable voice. The inspected ElevenLabs TTS tool does not expose voice changing or V4/V4 Turbo dialogue routes; those require different endpoint integrations. They are not native model impossibilities. For provider-specific limits and proof levels, use the ElevenLabs reference.
How do I place narration at a timecode?
Prepare the words
Finalize the script and language. Keep each request within the chosen model’s documented ceiling.
Choose the voice
Preview and attach a suitable voice. Selection alone does not generate speech.
Specify placement
Ask for the completed narration on an available audio track starting at a timecode, such as 00:03. Review the estimate before submitting.
Review the result
Find the asset in My Assets. Add it manually if needed, then check pronunciation, pauses, volume and sync. Move or trim the audio clip.
Example request: “Use the attached voice to read this approved script calmly. Place it on an available audio track starting at 00:03: Your next idea starts here.” This describes the intended edit, not a completed generation test. The expected result is an editable audio clip after generation and placement succeed.
Which controls change delivery?
For the inspected ElevenLabs route, speed changes pace and stability changes the balance between consistency and variation. Use audio tags only when the selected voice and model support them. A target such as “about twenty seconds” is creative direction, not an exact-duration field. Generate, listen, then revise the words or performance and fit the asset to the picture.
If you need an existing recording cleaned instead, use voice isolation. Review captions and music as separate parts of the mix.
Frequently asked questions
How much text fits?
How do I place narration at a timecode?
How do usage and playback length differ?
Sources checked 2026-10-28: ChatCut voiceover guide, model and endpoint evidence, official models and TTS API. No matched narration quality benchmark is supplied. For failures, see workflow recovery.
