Support OpenRouter image generation

[Powerostad]

Question

Hi, I have openrouter account and trying to use the nano banana model via openrouter with pydantic_ai, the problem is that the model response is str and the image that is being generated by model is not in the response object. Can someone Help me?
sample code:

from pydantic_ai import Agent, BinaryImage
from pydantic_ai.models.openai import OpenAIModel
from pydantic_ai.providers.openrouter import OpenRouterProvider

def generate_image(prompt: str, api_key: str):
    """Generate image using pydantic_ai with OpenRouter."""
    
    # Configure model
    provider = OpenRouterProvider(api_key=api_key)
    model = OpenAIModel(
        model_name="google/gemini-2.5-flash-image-preview",  # Doesn't exist on OpenRouter
        provider=provider,
    )
    
    # Create agent expecting BinaryImage
    agent = Agent(model=model, output_type=BinaryImage)
    
    # This fails: model returns text instead of image data
    result = agent.run_sync(prompt)  # Error: "Exceeded maximum retries for output validation"
    
    return result.output.data

if __name__ == "__main__":
    try:
        image_data = generate_image("A cat", "your_api_key")
        print("Success!")
    except Exception as e:
        print(f"Error: {e}")

Additional Context

No response

[DouweM]

@Powerostad The OpenRouter API requires specific fields to be set to generate images: https://openrouter.ai/docs/features/multimodal/image-generation. Since image generation is not supported by the official OpenAI Chat Completions API, we haven't implemented support for these fieldds in the OpenAIChatModel you're using.

We can add it to the OpenRouterModel that's in development at #3089.

It would need to pass modalities in the request similar to how we do in GoogleModel:

pydantic-ai/pydantic_ai_slim/pydantic_ai/models/google.py

Lines 419 to 421 in a70c2a5

	modalities = [Modality.TEXT.value]
	if self.profile.supports_image_output:
	modalities.append(Modality.IMAGE.value)

And parse images from the response: https://openrouter.ai/docs/features/multimodal/image-generation