# MusicGen

> Meta's MusicGen via Replicate. Generates music from a text prompt or a reference melody (stereo / melody / large variants), with continuation of an input clip and MP3 / WAV output.

- **Provider**: replicate
- **Model ID**: meta/musicgen
- **Category**: music_generation
- **Credits**: from 94 for the default configuration — varies by input, see Pricing
- **Pricing Type**: formula

## API Endpoint

Base URL: https://api.core.today/v1

### Create Prediction
POST /predictions

### Get Status
GET /predictions/{job_id}

### Cancel
DELETE /predictions/{job_id}

## Authentication

Header: `X-API-Key: YOUR_API_KEY`

## Input Parameters

- `prompt` (string, **required**): A description of the music you want to generate.
- `continuation_end` (integer, optional): End time (s) of input_audio for continuation. Omit for the end of the clip. (Range: min: 0)
- `model_version` (string, optional): Model to use. stereo-* produce stereo output; melody-* accept input_audio as a melody reference. (Default: `stereo-melody-large`; Options: `stereo-melody-large`, `stereo-large`, `melody-large`, `large`)
- `seed` (integer, optional): Seed for the random number generator. Omit or -1 for random.
- `multi_band_diffusion` (boolean, optional): Decode EnCodec tokens with MultiBand Diffusion (non-stereo models only). (Default: `false`)
- `continuation_start` (integer, optional): Start time (s) of input_audio to use for continuation. (Default: `0`; Range: min: 0)
- `duration` (integer, optional): Duration of the generated audio in seconds (1-30). (Default: `8`; Range: min: 1, max: 30)
- `top_p` (number, optional): Nucleus sampling probability. 0 uses top_k sampling. (Default: `0`)
- `input_audio` (string, optional): An audio file that will influence the generated music. With continuation it is continued; otherwise its melody is mimicked (melody models only).
- `output_format` (string, optional): Output format for generated audio. (Default: `wav`; Options: `wav`, `mp3`)
- `top_k` (integer, optional): Reduces sampling to the k most likely tokens. (Default: `250`)
- `temperature` (number, optional): Sampling temperature. Higher means more diversity. (Default: `1`)
- `normalization_strategy` (string, optional): Strategy for normalizing audio. (Default: `loudness`; Options: `loudness`, `clip`, `peak`, `rms`)
- `classifier_free_guidance` (integer, optional): Influence of inputs on the output. Higher adheres more to the prompt. (Default: `3`)
- `continuation` (boolean, optional): If true, generated music continues from input_audio. (Default: `false`)

## Example Request

```json
{
  "model": "meta/musicgen",
  "input": {
    "duration": 8,
    "model_version": "stereo-large",
    "output_format": "mp3",
    "prompt": "Edo25 major g melodies that sound triumphant and cinematic, leading up to a crescendo that resolves in a 9th harmonic"
  }
}
```

## Response Format

```json
{
  "job_id": "abc123",
  "status": "pending",
  "provider": "replicate",
  "model": "meta/musicgen",
  "created_at": "2026-01-01T00:00:00Z",
  "result": null,
  "error": null,
  "credits_used": null
}
```

Status values: `pending`, `processing`, `completed`, `failed`, `cancelled`

When the job is `completed`, `result` holds the output URL(s) or data and `credits_used` the net credits charged. When it is `failed`, `error` and `error_code` say why and the charge is refunded.

## Usage Flow

1. POST /predictions with model and input → receive job_id
2. Poll GET /predictions/{job_id} until status is `completed` or `failed`
3. Result contains output URL(s) or data

## Pricing

- Formula: `base + rate * duration` credits
  - `base`: 30.5
  - `duration`: the `duration` input (default 8)
  - `rate`: 7.92

Default configuration: 94 credits.

These are list prices. The amount charged is rounded, and large charges are discounted progressively. `POST /predictions/estimate` with the same `model` and `input` returns the charge for that input without creating a job (an estimate when the price depends on the output).

## Output Type

url

## Tags

music-generation, text-to-music, melody-conditioning, instrumental, open-source, meta

## Documentation

https://replicate.com/meta/musicgen