Imagen 4 API Quickstart: Google's Latest Image Model with Python

TL;DR

• Imagen 4 has three tiers: Fast (3–5s, cheapest), Standard (balanced), Ultra (highest quality, 20–45s).
• Connect through CodeGateway's OpenAI-compatible endpoint — same SDK, just change base_url.
• strong for: ultra-realistic photos and human portraits. Does not support transparent backgrounds or inpainting.

Imagen 4 Tier Specs

Source: Google Cloud Vertex AI documentation (May 2026). Check Google's pricing page for current rates.

CodeGateway adds a tier multiplier (1.2x–1.5x). See CodeGateway pricing tiers.

Quality Benchmark

In May 2026 internal testing, we ran the same prompt through all three Imagen 4 tiers and GPT-Image-2 (high quality):

Prompt: "An Asian female software engineer in a modern office, holding a coffee cup, code on her monitors, cinematic lighting, ultra-realistic photography style"

Takeaway: Imagen 4 Ultra wins on photorealism and face fidelity. GPT-Image-2 holds the lead on text rendering accuracy.

Python Quickstart

pip install openai

from openai import OpenAI

# Imagen 4 via CodeGateway's OpenAI-compatible endpoint
client = OpenAI(
    api_key="your-codegateway-api-key",
    base_url="https://api.codegateway.dev/v1"
)

# Imagen 4 Standard — good default
response = client.images.generate(
    model="imagen-4-0",            # or "imagen-4-flash" / "imagen-4-ultra"
    prompt="A modern city skyline wrapped in morning fog, ultra-realistic photography, golden hour light, 4K detail",
    n=1,
    size="1024x1024"
)

print(f"Image URL: {response.data[0].url}")  # Valid ~60 minutes

Save to disk

import base64

response = client.images.generate(
    model="imagen-4-ultra",
    prompt="A young engineer with glasses, intensely focused on three monitors, realistic office environment, natural light",
    n=1,
    size="1024x1024",
    output_format="jpeg"
)

img_bytes = base64.b64decode(response.data[0].b64_json)
with open("output.jpg", "wb") as f:
    f.write(img_bytes)

aspectRatio Parameter

Imagen 4 supports aspect ratio control — a feature GPT-Image-2 lacks.

# Portrait (9:16) — mobile wallpapers, short video covers
response = client.images.generate(
    model="imagen-4-0",
    prompt="Cyberpunk city street, neon rain, vertical composition, cinematic",
    n=1,
    aspect_ratio="9:16"
)

# Landscape (16:9) — website banners, video thumbnails
response = client.images.generate(
    model="imagen-4-0",
    prompt="Minimalist white office, floor-to-ceiling windows, sunlight, commercial photography",
    n=1,
    aspect_ratio="16:9"
)

Supported values: 1:1, 4:3, 3:4, 16:9, 9:16, 3:2, 2:3.

Batch Generation

# Imagen 4 supports up to 4 images per request
response = client.images.generate(
    model="imagen-4-0",
    prompt="Professional API documentation screenshots, tech aesthetic, blue color scheme",
    n=4,
    size="1024x1024"
)

for i, img in enumerate(response.data):
    print(f"Variant {i+1}: {img.url}")

Tier Selection Guide

Cost Comparison (100 images/day, CodeGateway new user 1.5x)

Imagen 4 Fast is the budget option for high-volume, lower-stakes generation. Ultra and GPT-Image-2 high are in similar price territory but serve different use cases (photorealism vs. text rendering).

FAQ

Q: English or non-English prompts for Imagen 4?

A: A: English prompts give more precise control over style, lighting, and composition descriptors. Other languages work, but key style terms in English tend to produce more predictable results.

Q: How long are generated image URLs valid?

A: A: Approximately 60 minutes. For durable storage, request b64_json format and save to your own object storage.

Q: How much slower is Ultra vs. Standard?

A: A: Ultra: 20–45s. Standard: 8–15s. Fast: 3–5s. For any user-facing real-time display, use Fast or Standard.

Q: How does Imagen 4 compare to Gemini's image generation?

A: A: Imagen 4 is a dedicated image generation API optimized for high-quality output. Gemini's image generation is integrated into a multimodal model, better suited for conversational "generate an image while we chat" scenarios. For dedicated image generation tasks, Imagen 4 is the right tool.

Q: Does CodeGateway support both Imagen 4 and GPT-Image-2?

A: A: Yes — both models work through the same CodeGateway API key and base_url. Switch between them by changing the model parameter.

Related

• Image API Comparison: Imagen, Gemini, and GPT-Image
• GPT-Image-2 API Quickstart
• CodeGateway Pricing Tiers
• CodeGateway Billing Guide
• Google Imagen Documentation

References

Anthropic multimodal docs · OpenAI image generation reference