# openrouter-vision

> Any LLM is a versatile large language model for text generation, comprehension, and diverse NLP tasks such as chat and summarization. Ready-to-use REST inference API, best performance, no coldstarts, affordable pricing.

## Overview

- **Endpoint**: `POST https://api.muapi.ai/api/v1/openrouter-vision`
- **Model ID**: `openrouter-vision`
- **Category**: text to text
- **Variant**: Image to Text
- **Family**: llm
- **Cost**: 0.025 credits per call (some models compute cost dynamically based on params)

## API Usage

MuApi uses a **submit-then-poll** pattern: submit a job, get a `request_id`, then poll the predictions endpoint until `status` is `completed`. Optionally pass `?webhook=YOUR_URL` on the submit call to receive a POST callback when the job finishes (skip polling).

**Authentication**: send your MuApi key in the `x-api-key` header. Get one at https://muapi.ai/access-keys.

### 1. Submit a job

```http
POST https://api.muapi.ai/api/v1/openrouter-vision
Content-Type: application/json
x-api-key: YOUR_API_KEY
```

**Minimum (required only):**

```json
{
  "prompt": "<prompt>",
  "images_list": "https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/openrouter-vision.jpg"
}
```

**Full example (all params):**

```json
{
  "prompt": "<prompt>",
  "images_list": "https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/openrouter-vision.jpg",
  "system_prompt": "<system_prompt>",
  "model": "google/gemini-2.5-flash",
  "reasoning": false,
  "temperature": 1,
  "max_tokens": null
}
```

**Response:**

```json
{
  "request_id": "abc123",
  "status": "processing"
}
```

### 2. Poll for the result

```http
GET https://api.muapi.ai/api/v1/predictions/{request_id}/result
x-api-key: YOUR_API_KEY
```

Possible `status` values: `queued`, `pending`, `processing`, `completed`, `failed`, `cancelled`. Poll every 2-5 seconds until terminal. When `completed`, the result URLs are in the `outputs` array.

**Example response when `completed`:**

```json
{
  "id": "abc123",
  "status": "completed",
  "outputs": [
    "https://cdn.muapi.ai/.../output.png"
  ],
  "urls": {
    "get": "https://api.muapi.ai/api/v1/predictions/abc123/result"
  },
  "created_at": "2026-05-08T12:34:56Z",
  "has_nsfw_contents": []
}
```

### cURL

```bash
# 1. Submit
REQUEST_ID=$(curl -s -X POST https://api.muapi.ai/api/v1/openrouter-vision \
  -H "x-api-key: $MUAPI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"<prompt>","images_list":"https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/openrouter-vision.jpg"}' | jq -r .request_id)

# 2. Poll until completed
while :; do
  RESP=$(curl -s https://api.muapi.ai/api/v1/predictions/$REQUEST_ID/result -H "x-api-key: $MUAPI_API_KEY")
  STATUS=$(echo "$RESP" | jq -r .status)
  [ "$STATUS" = "completed" ] && echo "$RESP" | jq .outputs && break
  [ "$STATUS" = "failed" ] && echo "$RESP" && exit 1
  sleep 3
done
```

### Python

```python
import os, time, requests

API = "https://api.muapi.ai/api/v1"
headers = {"x-api-key": os.environ["MUAPI_API_KEY"]}

r = requests.post(f"{API}/openrouter-vision", headers=headers, json={"prompt":"<prompt>","images_list":"https://d3adwkbyhxyrtq.cloudfront.net/webassets/videomodels/openrouter-vision.jpg"})
request_id = r.json()["request_id"]

while True:
    res = requests.get(f"{API}/predictions/{request_id}/result", headers=headers).json()
    if res["status"] == "completed":
        print(res["outputs"]); break
    if res["status"] == "failed":
        raise RuntimeError(res.get("error")); 
    time.sleep(3)
```

## Input Schema

The API accepts the following input parameters:

- **`prompt`** (`string`, _required_):
  The prompt to generate the response

- **`images_list`** (`array`, _required_):
  Upload or provide image urls. Used for image-to-video generation.

- **`system_prompt`** (`string`, _optional_):
  System prompt to provide context or instructions to the model.

- **`model`** (`string`, _optional_):
  Name of the model to use. Premium models are charged at 10x the rate of standard models, they include: deepseek/deepseek-r1, google/gemini-pro-1.5, openai/gpt-4.1, anthropic/claude-3-5-haiku, openai/gpt-4o, anthropic/claude-3.5-sonnet, openai/o3, meta-llama/llama-3.2-90b-vision-instruct, anthropic/claude-3.7-sonnet, openai/gpt-5-chat.
  - Default: `"google/gemini-2.5-flash"`
  - Options: `"google/gemini-2.5-flash"`, `"anthropic/claude-sonnet-4.5"`, `"openai/gpt-4o"`, `"qwen/qwen3-vl-235b-a22b-instruct"`, `"x-ai/grok-4-fast"`

- **`reasoning`** (`boolean`, _optional_):
  Should reasoning be the part of the final answer.
  - Default: `false`

- **`temperature`** (`int`, _optional_):
  This setting influences the variety in the model's responses. Lower values lead to more predictable and typical responses, while higher values encourage more diverse and less common responses. At 0, the model always gives the same response for a given input.
  - Default: `1`

- **`max_tokens`** (`int`, _optional_):
  This sets the upper limit for the number of tokens the model can generate in response. It won’t produce more than this limit. The maximum value is the context length minus the prompt length.
  - Default: `null`

## Output Schema

The polling endpoint returns the following fields:

- **`id`** (`string`): The request ID.
- **`status`** (`string`): One of `queued`, `pending`, `processing`, `completed`, `failed`, `cancelled`.
- **`outputs`** (`array`): URLs to generated images/videos/audio. Empty until `status` is `completed`.
- **`urls.get`** (`string`): Self-link to re-fetch this prediction.
- **`error`** (`string` | `null`): Error message if `status` is `failed`.
- **`created_at`** (`string`): ISO-8601 timestamp of when the request was created.
- **`has_nsfw_contents`** (`array of boolean`): Per-output NSFW detection flags.

## Webhooks (optional)

Append `?webhook=https://your-server/path` to the submit URL. When the job reaches a terminal state, MuApi will POST the same shape as the polling response to your URL — no polling needed.

## Agent Integration

MuApi ships an MCP server and CLI so agents (Claude Code, Cursor, custom) can call this endpoint without writing HTTP code:

```bash
# Install the CLI
npm install -g muapi-cli

# Authenticate once
muapi auth login

# Expose all MuApi models as MCP tools to your agent
muapi mcp serve
```

The MCP server exposes tools that wrap submit + poll for every model, including `openrouter-vision`. See `muapi --help` for category-specific shortcuts (`muapi image generate`, `muapi video from-image`, etc.).

## Related Models
- [Text to Text](https://muapi.ai/playground/any-llm)
- [Image to Text](https://muapi.ai/playground/openrouter-vision)

## Resources
- [Playground Page](https://muapi.ai/playground/openrouter-vision)
- [API Reference](https://muapi.ai/playground/openrouter-vision?tab=2)
- [Global llms.txt](https://muapi.ai/llms.txt)