Gemini Image Generation with LangChain

main.py

from langchain_google_genai import ChatGoogleGenerativeAI, Modality
from base64 import b64decode
from datetime import datetime
from uuid import uuid4

model: str = "gemini-3-pro-image-preview"
# Force modality to discard verbal chat answer
llm = ChatGoogleGenerativeAI(model=model, response_modalities=[Modality.IMAGE])

prompt = "Generate an image of a Pinscher in a suit, on top of a table in a club. Shot with iPhone 6."
response = llm.invoke(prompt)

image_b64 = response.content[0]["image_url"]["url"]
# Convert b64 string to bytes
header, encoded = image_b64.split(",", 1)
image_bytes = b64decode(encoded)

filename = f"{str(uuid4())}.png"

with open(filename, "wb") as f:
    f.write(image_bytes)
    print(f"[INFO] File {filename} written to disk.")