---
name: reels-voice
description: "Choose and keep the voice of an AI character for a Reels / Shorts series — ElevenLabs v4 (best), natural voices from the ElevenLabs Voice Library, a voice designed from a text description (Qwen3-TTS voice-design or MiniMax voice-design). Timings and source picks. Part of the reels reference set."
---

These creative references describe source experiments with external tools. Use your own supported account and tools; this download installs no automation. Settings and timings are experimental observations, not current FlashBoards availability. Use only media and voices you have permission to use.

# reels-voice · the character's voice

Pick {voice} for {character}: their own clone if they have one (`reels-voice-clone`), otherwise a library or designed voice that fits them. Tested 9.10.2026 on one 85-character line (see the example run). Every file checked with speech-to-text: all words right, no tag read aloud.

## Choose

| Priority | Model · voice | Time | Verdict |
|---|---|---|---|
| 🏆 **owner's pick** | **ElevenLabs v4** (`eleven_v4`) — direct API or `elevenlabs/tts/eleven-v4` on fal | 4–15 s | the most natural; tags work |
| 🏆 natural, ready | Voice Library, filter by gender, `use_cases` conversational / social_media, sorted by 1-year usage. Options: **Hope — Bubbly, Gossipy and Girly** (`uYXf8XasLslADfZ2MB4u`, female, natural "mmm", "ah", chuckles); **Donovan — Articulate, Strong and Deep** (`DMyrgzQFny3JI1Y1paM5`, male); **Spence — Clear, Soft-Spoken and Natural** (`ZTho75k1M56OV0k9XtSC`, male); **J Thomas** (`f5gk6ENs9vgDBvfkTj7e`, male, warm and raspy, lively) | — | options, not a default — match the person |
| 💰 own voice from a description | **Qwen3-TTS voice-design 1.7B** (`fal-ai/qwen-3-tts/voice-design/1.7b`) | 40 s | owner: good; the voice is not stored — clone the result (`reels-voice-clone`) |
| own voice, stored | MiniMax voice-design (`fal-ai/minimax/voice-design`) | 15 s | returns `custom_voice_id`, then reuse it with MiniMax Speech |
| ⚡💰 premade | ElevenLabs v4 Turbo on fal (`elevenlabs/tts/eleven-v4-turbo`): Charlotte, Alice, Lily, Jessica… | 3.6 s | fine, more "announcer" |

## Steps

1. Pick {voice}; keep its ElevenLabs `voice_id` (library voice or the person's clone id) in the person's character card / memory.
2. Voice every line with the same voice and model (`eleven_v4`) through your supported ElevenLabs tool. Supply the saved `voice_id`; a library identifier can also be used directly.
3. Check with speech-to-text before the video.

## Notes

- ElevenLabs on fal has **no voice cloning** — only premade voice names; clone directly in ElevenLabs.
- v4 Turbo is the real-time model (agents); for content use v4.

## Example run (reference only — never reuse its files, prompts or images for a person; select a permissioned voice)

The library voice names and identifiers above and below are external sample settings, not private credentials or a mandatory voice selection. Use only voices deliberately selected and permissioned through your own account; no example audio or character files are delivered or required.

- Demo character Mara Vale, «Mara's Rules» ep. 1, 9.10.2026. Line: "Rule number one: never go back for what you left behind. Rule number two... tomorrow." (85 characters).
- Series voice chosen for Mara: Hope (`uYXf8XasLslADfZ2MB4u`). Donovan and Spence were registered for the demo partner Dex (film-scene tests; Dex = Spence there).