Create Image

Image Generation with Azure OpenAI

This skill generates images using Azure OpenAI GPT Image models through direct API calls.

When to use

• User asks to generate, create, or produce an image
• User wants to visualize a concept or idea
• User requests artwork, illustrations, or graphics
• User asks to create visual representations of descriptions

Prerequisites

User must have:

1. Azure OpenAI resource with deployed GPT Image model (gpt-image-2 or gpt-image-1.5)
2. Azure authentication configured:
   • Recommended: Azure CLI (az login) with "Cognitive Services OpenAI User" role
   • Alternative: API key (if enabled on the endpoint)

Important: Many Azure OpenAI endpoints disable API key authentication. Always use Azure RBAC when API keys are disabled.

How to use

Using Python

Install required packages:

pip install openai azure-identity

Generate image with Azure RBAC (recommended):

from openai import AzureOpenAI
from azure.identity import DefaultAzureCredential, get_bearer_token_provider

token_provider = get_bearer_token_provider(
    DefaultAzureCredential(),
    "https://cognitiveservices.azure.com/.default"
)

client = AzureOpenAI(
    api_version="2024-02-01",
    azure_endpoint="https://YOUR-RESOURCE.openai.azure.com/",
    azure_ad_token_provider=token_provider
)

result = client.images.generate(
    model="gpt-image-2",
    prompt="A serene mountain landscape at sunset",
    size="1536x1024",
    quality="medium"
)

print(result.data[0].url)

Parameters

• prompt: Description of the image (1-4000 chars, required)
• model: gpt-image-2 or gpt-image-1.5 (use deployment name)
• size: Use model-supported dimensions such as 1024x1024, 1024x1536, or 1536x1024 (default: 1024x1024)
• quality: low, medium, high, or auto for GPT Image models (use medium by default)
• n: Number of images (model-specific limits apply)
• style: Optional stylistic control if supported by the deployed model

Prompt crafting tips

1. Be specific and descriptive
2. Include style, mood, and atmosphere details
3. Specify key visual elements and composition
4. Use portrait (1024x1536) for vertical, landscape (1536x1024) for horizontal images
5. Use high quality only when the user specifically requests extra detail; use low for faster or lower-cost drafts

Output

• The API returns a temporary URL valid for 24 hours
• Download and save the image if persistence is needed
• The response includes a revised_prompt showing what the model actually used

Error handling

• 401/403: Check Azure authentication and RBAC role assignments
• API Key Disabled: Switch to Azure RBAC with DefaultAzureCredential
• 429: Rate limit reached, implement retry with backoff
• Content filtered: Revise prompt to comply with content policies

References

• Azure OpenAI Image Generation Docs