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

# Generate video from audio

> Use a source audio track to drive an LTX video generation job.

Use this recipe to generate a video whose timing and performance follow an
existing audio track. This is useful for dialogue, narration, music, and other
audio-led scenes; it does not transcribe the audio or create a talking avatar by
itself.

## 1. Choose an audio-capable model

Audio-to-video is available on these LTX routes:

* `ltx-2-5-fast`
* `ltx-2-5-pro`
* `ltx-2-3-pro`
* `ltx-2-pro` until its published retirement date

Check `GET /v1/videos/models` before production requests because model status
and supported parameters can change.

## 2. Send the audio input

Provide exactly one `audio_url` reference with the `source` role. Set
`input_audio_duration` to the source audio length in seconds so Phaseo can
validate it against the model's supported range. Because LTX does not return
authoritative duration or usage, LTX audio-to-video jobs are currently reserved
and settled at the 20-second input ceiling.

```bash theme={null}
curl https://api.phaseo.app/v1/videos \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ltx-2-5-pro",
    "prompt": "A presenter delivers the supplied narration naturally in a modern studio.",
    "input_audio_duration": 8.5,
    "input_references": [
      {
        "type": "audio_url",
        "role": "source",
        "media_url": {
          "url": "https://example.com/narration.mp3"
        }
      }
    ]
  }'
```

The source must be a publicly reachable HTTPS URL. LTX audio inputs must be
between 2 and 20 seconds.

## 3. Optionally provide a first frame

Add one `image_url` reference with the `first_frame` role when you want to set
the opening appearance. Do not add a last frame to an LTX audio-driven request.

```json theme={null}
{
  "type": "image_url",
  "role": "first_frame",
  "image_url": {
    "url": "https://example.com/presenter.jpg"
  }
}
```

## 4. Poll or subscribe for completion

The create call returns an async video job. Poll its status URL, or add a
`webhook` configuration to receive lifecycle events. Download the output only
after the job reaches `completed`.

## Related guides

* [Choose a video generation model](./choose-a-video-generation-model.mdx)
* [Use image inputs for video generation](./use-image-inputs-for-video-generation.mdx)
* [Run async video with webhooks](./async-video-webhooks.mdx)
* [API Reference: Video generation](../api-reference/endpoint/video-generation.mdx)