Aspect Ratio In Image Generation Gemini

[laurasaarsaar]

Gemini Model Aspect Ratio Behavior In agents

I want to put a ascpetc ratio for image generaion using Gemini but there is no doc for this configuration in model of agents

Current Implementation

In our code, we're setting the aspect ratio in two places:

1. Scene Integration Agent:

from agno.agent import Agent
 Agent(
            tools=[self.generate_story_image],
            model=Gemini(
                id="gemini-2.5-flash-image",
                response_modalities=["Text", "Image"],
                api_key=settings.GOOGLE_API_KEY.get_secret_value(),
                temperature=0.70,
            )
        )

As in api call of google we have it
2. Image Generation:

config=types.GenerateContentConfig(
    temperature=0.0,
    candidate_count=1,
    image_config=types.ImageConfig(
        aspect_ratio="1:1",
    )
)

Issue

• The Gemini model's behavior regarding aspect ratio is unclear
• We're unsure whether aspect_ratio should be set in the model initialization or in the generation config
• Documentation doesn't clearly specify the precedence when both are set
• Different aspect ratios might be needed for different types of generations (avatars vs scenes)

Questions for Discussion

1. Where is the most effective place to set aspect ratio?
2. Does the model respect both settings or does one override the other?

Please share your experiences and observations with different aspect ratio settings in the Gemini model.

[ghost]

Quick Summary: Gemini aspect_ratio Configuration
Here are the direct answers to your questions:

Where is the most effective place to set aspect_ratio? In the generation config (types.GenerateContentConfig) at the moment of the API call. This should be done inside the logic of your tool.

Does one setting override the other? Yes. The generation config always overrides any default set on the model during initialization.

The correct pattern is to define the aspect_ratio for each specific task, giving your agent the flexibility to create different image types (e.g., 1:1 for avatars, 16:9 for scenes) using the same model instance.

Example Code Pattern
This shows how each tool should define its own configuration.

from google.generativeai import types

from your_gemini_client import gemini_model

def generate_avatar(prompt: str):
"""Generates a 1:1 square avatar."""
# Configuration is defined here, per-task
generation_config = types.GenerateContentConfig(
image_config=types.ImageConfig(aspect_ratio="1:1")
)

# In your real code, you would make the API call here
# response = gemini_model.generate_content(prompt, generation_config=generation_config)
print(f"API call for an avatar with aspect_ratio='1:1'")

def generate_scene(prompt: str):
"""Generates a 16:9 wide scene."""
# Configuration is defined here, per-task
generation_config = types.GenerateContentConfig(
image_config=types.ImageConfig(aspect_ratio="16:9")
)

# In your real code, you would make the API call here
# response = gemini_model.generate_content(prompt, generation_config=generation_config)
print(f"API call for a scene with aspect_ratio='16:9'")

--- Agent Usage Example ---

When the agent needs an avatar, it calls this tool:

generate_avatar("A close-up of a robot's face")

When the agent needs a scene, it calls this tool:

generate_scene("A futuristic city at sunset")

Conclusion: Always set aspect_ratio in the generation config for maximum flexibility. The model initialization in your Agent is correct as it is—it should not contain aspect ratio details.