Google Image Generation

Generate an image with Gemini response modalities and open it with PIL.

image_generation.py
"""
Google Image Generation
=======================

Cookbook example for `google/gemini/image_generation.py`.
"""

from io import BytesIO

from agno.agent import Agent, RunOutput  # noqa
from agno.models.google import Gemini
from PIL import Image

# ---------------------------------------------------------------------------
# Create Agent
# ---------------------------------------------------------------------------

# No system message should be provided
agent = Agent(
    model=Gemini(
        id="gemini-3.7-flash",
        response_modalities=["Text", "Image"],
    )
)

# Print the response in the terminal
run_response = agent.run("Make me an image of a cat in a tree.")

if run_response and isinstance(run_response, RunOutput) and run_response.images:
    for image_response in run_response.images:
        image_bytes = image_response.content
        if image_bytes:
            image = Image.open(BytesIO(image_bytes))
            image.show()
            # Save the image to a file
            # image.save("generated_image.png")
else:
    print("No images found in run response")

# ---------------------------------------------------------------------------
# Run Agent
# ---------------------------------------------------------------------------
if __name__ == "__main__":
    pass

Run the Example

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U agno google-genai pillow

Export your Google API key

export GOOGLE_API_KEY="your_google_api_key_here"

Use an image-generation model

In the saved source, replace id="gemini-3.7-flash" with id="gemini-3.1-flash-image". The original text-output model does not generate images. See native Gemini image generation.

Run the example

Save the code above as image_generation.py, then run:

python image_generation.py

Full source: cookbook/90_models/google/gemini/image_generation.py