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

# Image input (vision)

> How to send an image to an LLM via RuAPI: a Python example with image_url, public URL and base64, which models and formats are supported.

Vision-capable models take images alongside text in a single request. You can hand the model a picture and ask it to describe the scene, read text from it (OCR), make sense of a UI screenshot, a diagram or a chart, or pull data out of a photographed table.

Under the hood this uses the same OpenAI-compatible endpoint as a normal chat. The only thing that changes is the `content` field: instead of a string you pass an **array of parts** — text blocks `{"type": "text"}` and image blocks `{"type": "image_url"}`.

<Tip>
  Use a model that **supports vision**: [Claude](/en/claude-api) (Sonnet / Opus), [GPT](/en/gpt-api), [Gemini](/en/gemini-api) and [GLM](/en/glm-api) (GLM-5V) accept images. DeepSeek is text-only — see [DeepSeek models](/en/deepseek-api). Check the capability badges on the **Pricing** page at [www.ruapi.ai](https://www.ruapi.ai) before you send a request.
</Tip>

## Image from a public URL

The simplest case — the image is already online at a direct URL.

```python Python (OpenAI SDK) theme={null}
from openai import OpenAI

client = OpenAI(
    api_key="sk-YOUR_KEY",
    base_url="https://www.ruapi.ai/v1",
)

response = client.chat.completions.create(
    model="claude-opus-4-8",  # a vision model; exact names on the pricing page
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What's in this image? Describe it briefly."},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://upload.wikimedia.org/wikipedia/commons/3/3a/Cat03.jpg"
                    },
                },
            ],
        }
    ],
)

print(response.choices[0].message.content)
```

## Local file via base64

If the image isn't publicly reachable (a screenshot, a photo on disk), encode it to base64 and pass it as a **data URI** in the form `data:image/jpeg;base64,...`.

```python Python (OpenAI SDK) theme={null}
import base64
from openai import OpenAI

client = OpenAI(
    api_key="sk-YOUR_KEY",
    base_url="https://www.ruapi.ai/v1",
)

# Encode a local file to base64
with open("screenshot.jpg", "rb") as f:
    b64 = base64.b64encode(f.read()).decode("utf-8")

data_uri = f"data:image/jpeg;base64,{b64}"

response = client.chat.completions.create(
    model="claude-opus-4-8",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Read all the text in this screenshot."},
                {"type": "image_url", "image_url": {"url": data_uri}},
            ],
        }
    ],
)

print(response.choices[0].message.content)
```

<Note>
  For PNG change the prefix to `data:image/png;base64,`, for WebP use `data:image/webp;base64,`. The MIME type must match the file's real format.
</Note>

## Gotchas

<AccordionGroup>
  <Accordion title="Bigger images cost more tokens">
    An image counts as input tokens, and the higher the resolution the more expensive the request. If you don't need the fine detail, downscale the image before sending (for example to 1024–1568 px on the long side).
  </Accordion>

  <Accordion title="Huge base64 payloads bloat the request">
    A base64 string is about a third larger than the file itself and travels entirely inside the request body. For heavy images a public URL is safer than a multi-megabyte data URI — otherwise you can hit the request size limit.
  </Accordion>

  <Accordion title="Supported formats">
    Typically `jpeg`, `png`, `webp` and `gif`. Convert exotic formats (HEIC, TIFF, SVG) to one of these beforehand.
  </Accordion>
</AccordionGroup>

## What's next

* [Quick Start](/en/quickstart) — sign-up, key and your first request
* [Errors and how to fix them](/en/errors) — what to do on 400 / 413 / 415 and other codes