Generative AI Google Brain

gemini-getting-started-intro-gemini-3-image-gen

4 min read

Gemini 3 Pro Image (Nano Banana Pro 🍌) Generation

Source notebook

Repo path: gemini/getting-started/intro_gemini_3_image_gen.ipynb · Open on GitHub · intro

Shows how to generate and edit images with Gemini 3 Pro Image using the Google GenAI SDK.

Summary

This notebook teaches how to use Nano Banana Pro, loaded as gemini-3-pro-image, on Agent Platform with the Google Gen AI SDK. It walks through text-to-image generation, model thoughts, Google Search grounding, aspect ratios, image sizes, localization, multi-turn image editing, and multiple reference images. The workflow sets up a project client, calls generate_content with image configs, inspects response parts, and passes image bytes or Cloud Storage URIs for editing.

Key code patterns

Enterprise client setup

PROJECT_ID = os.environ.get("GOOGLE_CLOUD_PROJECT")
LOCATION = "global"
client = genai.Client(
    enterprise=True,
    project=PROJECT_ID,
    location=LOCATION,
)

Creates the Google GenAI SDK client for Agent Platform using a project and global location.

Text-to-image generation

response = client.models.generate_content(
    model="gemini-3-pro-image",
    contents=prompt,
    config=types.GenerateContentConfig(
        response_modalities=["IMAGE", "TEXT"],
        image_config=types.ImageConfig(aspect_ratio="16:9"),
    ),
)

Uses response_modalities and ImageConfig to request generated image output.

Finish reason guard

if response.candidates[0].finish_reason != types.FinishReason.STOP:
    reason = response.candidates[0].finish_reason
    raise ValueError(f"Prompt Content Error: {reason}")

Checks for prompt or content errors when an image is not generated.

Display image parts

for part in response.candidates[0].content.parts:
    if part.thought:
        continue
    if part.inline_data:
        display(Image(data=part.inline_data.data, width=1000))

Iterates response parts and skips thought content before displaying image bytes.

Google Search grounding

google_search = types.Tool(google_search=types.GoogleSearch())
response = client.models.generate_content(
    model=MODEL_ID,
    contents=prompt,
    config=types.GenerateContentConfig(
        response_modalities=["TEXT", "IMAGE"],
        tools=[google_search],
    ),
)

Adds a Google Search tool so the generated response can be grounded in search text results.

Image editing from bytes

with open(starting_image, "rb") as f:
    image = f.read()
response = client.models.generate_content(
    model=MODEL_ID,
    contents=[
        types.Part.from_bytes(data=image, mime_type="image/png"),
        "Change the text in this infographic from English to Spanish.",
    ],
)

Passes an input image and text instruction for localization through image editing.

Multi-turn image editing

chat = client.chats.create(
    model=MODEL_ID,
    config=types.GenerateContentConfig(response_modalities=["TEXT", "IMAGE"]),
)
response = chat.send_message(message)
response = chat.send_message([
    types.Part.from_bytes(data=data, mime_type="image/png"),
    "Make the perfume bottle purple.",
])

Keeps a chat session and feeds prior image bytes into the next edit request.

Multiple reference images

response = client.models.generate_content(
    model=MODEL_ID,
    contents=[
        types.Part.from_uri(file_uri="gs://cloud-samples-data/generative-ai/image/woman.jpg", mime_type="image/jpeg"),
        types.Part.from_uri(file_uri="gs://cloud-samples-data/generative-ai/image/suitcase.png", mime_type="image/png"),
        prompt,
    ],
)

Uses Cloud Storage URIs as multiple reference images for a composed generation.

Models & APIs used

  • Models: gemini-3-pro-image
  • APIs / services: Agent Platform API, Cloud Storage, Google Search
  • SDKs / libraries: google-genai, IPython, matplotlib, requests, Pillow

When to use this

Use this pattern when building Gemini image generation and editing workflows that need text prompts, search grounding, iterative edits, or reference images.

Gotchas & caveats

  • The notebook requires an existing Google Cloud project and the Agent Platform API enabled.
  • Colab users must authenticate with google.colab.auth.authenticate_user().
  • The client uses LOCATION = “global”.
  • To generate images, IMAGE must be included in response_modalities.
  • Valid aspect ratios are limited to 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, and 21:9.
  • Supported image sizes are 1K, 2K, and 4K.
  • Google Search grounding is based on text results, not images from Google Search.
  • The notebook notes higher latency than Gemini 2.5 Flash Image.
  • All generated images include a SynthID watermark.

Best practices

  • Install or upgrade google-genai before running the notebook.
  • Fallback to the GOOGLE_CLOUD_PROJECT environment variable when PROJECT_ID is not set.
  • Use GenerateContentConfig with response_modalities and ImageConfig for image output control.
  • Check finish_reason before assuming an image was generated.
  • Skip thought parts when displaying final generated images.
  • Display grounding citations and sources from grounding_metadata.
  • Use Part.from_bytes for local image editing inputs.
  • Use Part.from_uri for Cloud Storage reference images.

Graph View

Backlinks