# Voice first, then the face

Edit / Sound / Voice. Checked 2026-11-13.

The mouth is generated for the voice that ships, not for one that is replaced later.

- Use when: Generated people who speak: testimonials, ads, characters.
- Phrase: generate the final voice line first, with its delivery, then give that audio to the video model as the audio reference for the face
- Page: https://directorchair.ai/sound/voice-first
- JSON: https://directorchair.ai/api/sound/voice-first.json

## Watch out

- Not yet run by us. The order comes from a vendor launch whose film is a showreel of chosen takes; no failure rate is given. (practitioner; Director Chair notes on Chimera (ElevenLabs, Eleven v4 with Seedance 2.5))
- A voice reference used directly in the video model is often inconsistent between takes; the vendor's post and our own skill agree on that. (practitioner; Director Chair notes on Chimera (ElevenLabs, Eleven v4 with Seedance 2.5))
- A cloned voice speaking another language than its source loses its accent by design. Test before relying on an accent. (practitioner; Director Chair notes on Chimera (ElevenLabs, Eleven v4 with Seedance 2.5) (the vendor's own documentation, read 22 Oct 2026))
- Use your own voice or one with written consent. (practitioner; Director Chair notes on Chimera (ElevenLabs, Eleven v4 with Seedance 2.5))

## Prompts

### Any chat model

```
Give me a sound brief: what to record or pick, where it sits on the timeline, the levels, and what to check before export, for a short video about [your product], for [your audience]. The technique is voice first, then the face: the mouth is generated for the voice that ships, not for one that is replaced later. How it works: generate the final voice line first, with its delivery, then give that audio to the video model as the audio reference for the face. It is right for: generated people who speak: testimonials, ads, characters. Not yet run by us. A voice reference used directly in the video model is often inconsistent between takes; the vendor's post and our own skill agree on that. Be concrete: numbers, seconds and settings, no theory. Ask me one question first if something you need is missing.
```

## Goes well with

- Voice line: https://directorchair.ai/character/voice-line
- Prompt guide: Seedance 2.5: https://directorchair.ai/guide/guide-seedance-2-5
- Voice-over written to the cut: https://directorchair.ai/sound/voice-written-to-length

## Source

Own words from Director Chair notes on Chimera (ElevenLabs, Eleven v4 with Seedance 2.5)

Text CC BY 4.0, Director Chair (directorchair.ai). Reference clips and their card texts belong to their makers.