Advanced Gemini Image Generation Guide

This guide covers advanced scenarios for Gemini image generation including thinking process, Google Search grounding, and multi-turn conversations.

Thinking Process

The model uses internal reasoning for complex prompts. It generates interim thought images to refine composition before producing the final output.

Accessing Thoughts

uv run - << 'EOF'
# /// script
# dependencies = ["google-genai", "pillow"]
# ///
from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Create a detailed architectural blueprint of a futuristic city"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE']
    )
)

for part in response.parts:
    if part.thought:
        if part.text:
            print(f"Thought: {part.text}")
        elif image := part.as_image():
            image.save("tmp/thought.png")
            print("Thought image saved")
    elif part.inline_data is not None:
        part.as_image().save("tmp/output.png")
        print("Saved: tmp/output.png")
    elif part.text:
        print(part.text)
EOF

Thought Signatures

Thought signatures are encrypted representations of the model's internal thought process. They preserve reasoning context across multi-turn interactions for consistent iterative refinement.

Google Search Grounding

Generate images based on real-time information like weather forecasts, stock data, or current events.

Basic Search Grounding

uv run - << 'EOF'
# /// script
# dependencies = ["google-genai", "pillow"]
# ///
from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Visualize current weather for San Francisco as a chart"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
        tools=[{"google_search": {}}]
    )
)

for part in response.parts:
    if part.text:
        print(part.text)
    elif part.inline_data is not None:
        part.as_image().save("tmp/weather_chart.png")
        print("Saved: tmp/weather_chart.png")

# Access grounding metadata if available
if hasattr(response, 'grounding_metadata'):
    print(f"Sources: {response.grounding_metadata}")
EOF

Use Cases for Search Grounding

• Weather visualizations and forecasts
• Stock market charts and financial data
• Current events infographics
• Historical maps with accurate data
• Scientific diagrams requiring verified facts

Note: Image-based search results are not passed to the generation model.

Multi-Turn Conversations

Use chat sessions for iterative image refinement across multiple turns.

Chat-Based Editing

uv run - << 'EOF'
# /// script
# dependencies = ["google-genai", "pillow"]
# ///
from google import genai
from google.genai import types

client = genai.Client()

chat = client.chats.create(
    model="gemini-3-pro-image-preview",
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE']
    )
)

# First turn: Generate initial image
response = chat.send_message("Create an infographic explaining photosynthesis")

for part in response.parts:
    if part.inline_data is not None:
        part.as_image().save("tmp/infographic_v1.png")
        print("Saved: tmp/infographic_v1.png")

# Second turn: Modify the generated image
response = chat.send_message(
    "Translate all text to Spanish",
    config=types.GenerateContentConfig(
        image_config=types.ImageConfig(
            aspect_ratio="16:9",
            image_size="2K"
        )
    )
)

for part in response.parts:
    if part.inline_data is not None:
        part.as_image().save("tmp/infographic_v2.png")
        print("Saved: tmp/infographic_v2.png")
EOF

Multi-Turn Workflow Pattern

1. Generate - Create initial image
2. Review - Check the output
3. Refine - Send modification request in same chat
4. Iterate - Continue until satisfied

Multiple Reference Images

Combine up to 14 reference images (6 for objects, 5 for humans) to create composite images.

Group Photo Example

uv run - << 'EOF'
# /// script
# dependencies = ["google-genai", "pillow"]
# ///
from google import genai
from google.genai import types
from PIL import Image

client = genai.Client()

# Load reference images
person1 = Image.open("person1.png")
person2 = Image.open("person2.png")
person3 = Image.open("person3.png")

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        "Office group photo of these people making funny faces",
        person1,
        person2,
        person3,
    ],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
        image_config=types.ImageConfig(aspect_ratio="5:4")
    )
)

for part in response.parts:
    if part.inline_data is not None:
        part.as_image().save("tmp/group_photo.png")
        print("Saved: tmp/group_photo.png")
EOF

Professional Asset Production

Use gemini-3-pro-image-preview for high-fidelity professional outputs.

4K Marketing Asset

uv run - << 'EOF'
# /// script
# dependencies = ["google-genai", "pillow"]
# ///
from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Da Vinci style anatomical sketch of a butterfly"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
        image_config=types.ImageConfig(
            aspect_ratio="1:1",
            image_size="4K"
        )
    )
)

for part in response.parts:
    if part.text:
        print(part.text)
    elif part.inline_data is not None:
        part.as_image().save("tmp/butterfly_4k.png")
        print("Saved: tmp/butterfly_4k.png")
EOF

Text Rendering for Infographics

The pro model excels at generating legible, stylized text for:

• Infographics and data visualizations
• Restaurant menus and marketing materials
• Diagrams with labels
• Presentations and slides

Model Comparison