# Multimodal

> Use image, audio, video, and Realtime interfaces in LLMoxy.

Multimodal capabilities depend on the selected model. Confirm currently available multimodal models via `GET /v1/models` or the console pricing page.

## Images

OpenAI-style image generation:

```bash
curl https://llmoxy.com/v1/images/generations \
  -H "Authorization: Bearer <LLMOXY_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A clean product icon for an LLM gateway",
    "size": "1024x1024",
    "response_format": "url"
  }'
```

LLMoxy also forwards model-specific image parameters as long as the selected model supports them.

Image editing routes:

```txt
POST /v1/images/edits
POST /v1/edits
```

## Audio

Text-to-speech:

```bash
curl https://llmoxy.com/v1/audio/speech \
  -H "Authorization: Bearer <LLMOXY_API_KEY>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "tts-1",
    "input": "Hello from LLMoxy.",
    "voice": "alloy",
    "response_format": "mp3"
  }' \
  --output speech.mp3
```

Audio transcription and translation use multipart form upload:

```txt
POST /v1/audio/transcriptions
POST /v1/audio/translations
```

## Video

OpenAI-compatible video generation:

```bash
curl https://llmoxy.com/v1/videos \
  -H "Authorization: Bearer <LLMOXY_API_KEY>" \
  -F model=sora-2 \
  -F prompt="cute cat dance" \
  -F seconds=8
```

Query status:

```bash
curl "https://llmoxy.com/v1/videos/$TASK_ID" \
  -H "Authorization: Bearer <LLMOXY_API_KEY>"
```

Download content:

```bash
curl "https://llmoxy.com/v1/videos/$TASK_ID/content" \
  -H "Authorization: Bearer <LLMOXY_API_KEY>" \
  --output output.mp4
```

LLMoxy also supports the `POST /v1/video/generations` JSON video generation interface, as well as vendor-style routes such as Kling and Jimeng. See the console model list for specific supported models.

## Realtime

Realtime uses a WebSocket endpoint:

```txt
wss://llmoxy.com/v1/realtime?model=gpt-4o-realtime-preview
```

Server-side clients can use standard auth headers during the handshake; browser clients can use the OpenAI-style subprotocol key format.

This is the markdown version of https://en.llmoxy.com/docs/api-reference/multimodal.
