> ## Documentation Index
> Fetch the complete documentation index at: https://docs.musicmint.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Create AI Singer (Persona)

> Train a custom AI singer persona from an existing vocal clip, then use it to generate music with a consistent vocal style

## Endpoint

```http theme={null}
POST /api/v1/music/persona
```

Creates a reusable AI singer persona from an existing clip. Once created, pass the returned `persona_id` to [Generate Music](/api-reference/music/generate) to apply the vocal style to any new track.

## Request body

| Field     | Type   | Required | Description                                                    |
| --------- | ------ | -------- | -------------------------------------------------------------- |
| `name`    | string | **Yes**  | Display name for the persona (e.g. `"My Artist"` or `"我的歌手"`). |
| `clip_id` | string | **Yes**  | Source clip ID whose vocal style will be learned.              |

## Response

```json theme={null}
{
  "code": 0,
  "message": "ok",
  "request_id": "req-1710000000000",
  "data": {
    "taskBatchId": "batch-xyz789",
    "items": [
      {
        "id": "persona-abc123",
        "status": "complete"
      }
    ]
  }
}
```

| Field            | Type   | Description                                                       |
| ---------------- | ------ | ----------------------------------------------------------------- |
| `taskBatchId`    | string | Internal batch identifier                                         |
| `items`          | array  | Created persona items                                             |
| `items[].id`     | string | **Persona ID** — pass this as `persona_id` in generation requests |
| `items[].status` | string | `"complete"` when persona is ready to use                         |

## Using your Persona to generate music

Once you have a `persona_id`, pass it to any [Generate Music](/api-reference/music/generate) request. The AI will sing the generated track in the learned vocal style.

```bash theme={null}
curl -X POST https://api.example.com/api/v1/music/generate \
  -H "Authorization: Bearer sk-mm-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "[Verse]\n月光照在窗台\n风轻轻吹来\n[Chorus]\n你是我心中的光",
    "title": "心中的光",
    "tags": "流行 抒情 女声",
    "persona_id": "persona-abc123"
  }'
```

<Tip>
  `persona_id` works with both **custom mode** (`prompt` + `tags`) and **inspiration mode** (`gpt_description_prompt`). The vocal style is applied regardless of which generation mode you choose.
</Tip>

## Complete workflow

```
1. Record or generate a source clip with the target vocal style
        ↓
2. (Recommended) Run Stem Separation to isolate clean vocals
        ↓
3. POST /api/v1/music/persona  →  receive persona_id
        ↓
4. POST /api/v1/music/generate  +  persona_id  →  task_id
        ↓
5. Poll GET /api/v1/task/{task_id}  →  audio_url
```

## Create persona example

```bash theme={null}
curl -X POST https://api.example.com/api/v1/music/persona \
  -H "Authorization: Bearer sk-mm-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "My Custom Singer",
    "clip_id": "abc123def456"
  }'
```

## Choosing a source clip

For best results, use a clip with:

* Clear, isolated vocals (minimal background noise)
* At least 30 seconds of vocal content
* A consistent vocal style throughout

<Tip>
  Run [Stem Separation](/api-reference/audio/stem) on a clip first to get a clean vocal track, then use that vocal audio as the source for persona creation.
</Tip>