---
name: reels-voice-clone
description: "Clone a voice from a short recording and reuse it — Qwen3-TTS clone (best on fal, owner's pick), ElevenLabs Instant Voice Clone (direct API, voice_id kept), MiniMax voice-clone, Zonos2, Chatterbox HD; where the clone is stored and how to call it again. 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-clone · clone, save, reuse

Clone {voice} from the person's own recording (or from a designed voice). Tested 9.10.2026 on a 12.5-second sample (phone recording, .ogg → 44.1 kHz mono wav) and one 85-character line.

## Choose

| Priority | Model | Stored as | Time | Verdict |
|---|---|---|---|---|
| 🏆 **owner's pick on fal** | **Qwen3-TTS** `fal-ai/qwen-3-tts/clone-voice/1.7b` (+ `reference_text` of the sample) → `fal-ai/qwen-3-tts/text-to-speech/1.7b` with `speaker_voice_embedding_file_url` | a `.safetensors` speaker embedding — keep the file | 100 s embed + 23 s per line | closest to the source |
| 🏆 ElevenLabs quality | **ElevenLabs Instant Voice Clone** (`POST /v1/voices/add`) → TTS with `eleven_v4` | `voice_id` in your ElevenLabs account (takes 1 of 30 slots on Creator) | clone 3.5 s, line 5–15 s | owner: v4 best of v4 / v4 Turbo / v3 / Multilingual v2 |
| stored voice | MiniMax `fal-ai/minimax/voice-clone` (sample ≥ 10 s) | `custom_voice_id` | 11 s | good |
| no storage | Zonos2 `fal-ai/zonos2` · Chatterbox HD `resemble-ai/chatterboxhd/text-to-speech` | the sample goes with every request | 249 s / 132 s | weaker |

A voice can be **designed, then cloned**: make it with Qwen voice-design (`reels-voice`), then clone the result into ElevenLabs and voice on v4.

Whatever the method, keep the result (ElevenLabs `voice_id`, MiniMax `custom_voice_id`, or the Qwen `.safetensors` file + its `reference_text`) in the person's character card / memory.

## Steps (ElevenLabs)

1. In your own supported ElevenLabs account, create an Instant Voice Clone from a recording you have permission to use (`POST /v1/voices/add`).
2. Keep its `voice_id` with your character references.
3. Voice the line with that identifier and `model_id: eleven_v4`, then check the words.

## Rules

- Transcribe the sample first (Scribe v2) and pass it as `reference_text` to Qwen.
- Only clone voices you have the right to use.
- Slots: your own clones count (30 on Creator); library voices do not. Deleting a clone is permanent; a professional clone in verification cannot be deleted.

## Example run (reference only — never reuse its files, prompts, voices or images for a person)

The demo clone's identifier below is historical context, not a supplied voice or a reusable default. Replace it with your own authorized clone's `voice_id`; no example recording, clone or local runner is delivered or required.

- Demo character Mara Vale, 9.10.2026: voice made with Qwen voice-design , cloned into ElevenLabs as `mara` (`iV8DaEoP6jYNZUgTCz0e`), voiced on v4.