# Vision input

> Two ways to send images — URL or data URL (base64), plus size and format constraints.

Multimodal models accept image input. In the OpenAI protocol an image is an `image_url` content block.

## Public URL

```python
resp = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "What is in this image?"},
            {"type": "image_url", "image_url": {"url": "https://example.com/cat.jpg"}},
        ],
    }],
)
```

## base64 (data URL)

```python
import base64, pathlib

data = base64.b64encode(pathlib.Path("cat.jpg").read_bytes()).decode()
resp = client.chat.completions.create(
    model="gpt-5.4",
    messages=[{
        "role": "user",
        "content": [
            {"type": "text", "text": "Describe this image"},
            {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{data}"}},
        ],
    }],
)
```

## Things to watch

- **Format**: `image/jpeg`, `image/png` and `image/webp` are safest; SVG is usually rejected
- **Size**: base64 inflates the body by roughly a third — downscale so the long edge is 1024–2048px
- **Reachability**: the upstream vendor must be able to fetch your URL; private addresses and expired
  presigned links fail
- **Capability**: not every model accepts images; unsupported models return an error instead of
  ignoring the image

<Callout type="info" title="Anthropic and Gemini shapes">
With the Anthropic protocol an image is `{"type":"image","source":{"type":"base64",...}}`; with Gemini
it is `inline_data` / `file_data`. The gateway interprets the inbound protocol and maps it — see
[Protocol bridging](/en/guides/protocol-interop/).
</Callout>