openrouter-image-generation

name	openrouter-image-generation
description	Fix for OpenRouter image generation response parsing. Use when: (1) using OpenRouter API with image generation models like google/gemini-2.5-flash-image, (2) response doesn't contain expected image data, (3) looking for base64 image in wrong location. Images are returned in message.images array, not in standard OpenAI format. Covers correct endpoint, response parsing, and base64 extraction.
author	Claude Code
version	1.0.0
date	"2026-01-23T00:00:00.000Z"

OpenRouter Image Generation Response Format

Problem

When using OpenRouter's image generation models (like google/gemini-2.5-flash-image), the response format differs from expectations. Images are not returned in the standard location, causing parsing failures.

Context / Trigger Conditions

• Using OpenRouter API (https://openrouter.ai/api/v1/chat/completions) for image generation
• Model is an image generation model (e.g., google/gemini-2.5-flash-image)
• Response returns 200 OK but no image found in expected location
• Looking for image URL or base64 data and not finding it

Solution

1. Use the Chat Completions Endpoint (NOT /images/generations)

response = requests.post(
    "https://openrouter.ai/api/v1/chat/completions",  # NOT /images/generations
    headers={
        "Authorization": f"Bearer {OPENROUTER_API_KEY}",
        "Content-Type": "application/json",
    },
    json={
        "model": "google/gemini-2.5-flash-image",
        "messages": [{"role": "user", "content": "Generate a pixel art sword"}],
    },
)

2. Extract Image from Correct Location

Images are returned in choices[0].message.images array, NOT in a standard image field:

result = response.json()
message = result.get("choices", [{}])[0].get("message", {})
images = message.get("images", [])

if images:
    # Image URL is a base64 data URL
    img_url = images[0].get("image_url", {}).get("url", "")
    # Format: "data:image/png;base64,<base64_data>"

    if img_url.startswith("data:image"):
        # Extract base64 data after the comma
        b64_data = img_url.split(",")[1]
        img_bytes = base64.b64decode(b64_data)

3. Complete Response Structure

{
  "choices": [{
    "message": {
      "content": "Here is your image...",
      "images": [{
        "image_url": {
          "url": "data:image/png;base64,iVBORw0KGgo..."
        }
      }]
    }
  }]
}

Verification

• Check that response.json()["choices"][0]["message"]["images"] exists and has items
• Verify the image_url.url starts with data:image
• Decode base64 and write to file - should produce valid image

Example

import requests
import base64

def generate_image(prompt: str, api_key: str) -> bytes:
    response = requests.post(
        "https://openrouter.ai/api/v1/chat/completions",
        headers={
            "Authorization": f"Bearer {api_key}",
            "Content-Type": "application/json",
        },
        json={
            "model": "google/gemini-2.5-flash-image",
            "messages": [{"role": "user", "content": prompt}],
        },
        timeout=120,
    )

    result = response.json()
    images = result["choices"][0]["message"].get("images", [])

    if not images:
        raise ValueError("No image in response")

    img_url = images[0]["image_url"]["url"]
    b64_data = img_url.split(",")[1]
    return base64.b64decode(b64_data)

# Usage
img_bytes = generate_image("A pixel art treasure chest", api_key)
with open("treasure.png", "wb") as f:
    f.write(img_bytes)

Notes

• All models generate at fixed sizes (typically 1024x1024) regardless of requested size
• The /images/generations endpoint may return 404 for some models - use chat/completions instead
• Cost-effective models include google/gemini-2.5-flash-image for pixel art generation
• Include style instructions in the prompt for consistent results (e.g., "16-bit pixel art, SNES style")

References

• OpenRouter API Documentation: https://openrouter.ai/docs
• Available image models can be queried via: GET https://openrouter.ai/api/v1/models