Text-to-image generation turns a written description into a picture — useful for marketing assets, mockups, illustrations, or just exploring ideas without opening a design tool. With Microsoft Foundry, generating an image is a single API call against your project's OpenAI-compatible image endpoint.

What you need

1. A Foundry project with an image-capable model deployment (e.g. MAI-Image-2.6 or gpt-image-1).
2. Azure CLI login (az login) with the Azure AI User role on the project — or any identity DefaultAzureCredential can pick up. No API keys to manage.
3. The Python packages listed in code/requirements.txt:

pip install -r Code/requirements.txt

The core idea

Generating an image against a Foundry deployment boils down to four steps:

1. Authenticate — get a token provider Foundry trusts.
2. Choose the image route — MAI image deployments use /mai/v1/images/generations; Azure OpenAI image deployments use /openai/v1/images/generations.
3. POST the prompt — send the deployment name, prompt, and dimensions to the right Foundry route.
4. Decode and save — the response returns the image as base64, or occasionally as a temporary URL; save the bytes to a file.

import os
from pathlib import Path
from urllib.parse import urlsplit

from azure.identity import DefaultAzureCredential, get_bearer_token_provider
from dotenv import load_dotenv
import requests

load_dotenv(Path(".env"))

project_url = urlsplit(os.environ["PROJECT_ENDPOINT"])
origin = f"{project_url.scheme}://{project_url.netloc}"
model = os.environ["MODEL"]
prompt_file = os.getenv("PROMPT_FILE")
prompt = Path(prompt_file).read_text(encoding="utf-8").strip() if prompt_file else os.environ["PROMPT"]
width = int(os.getenv("WIDTH", "1024"))
height = int(os.getenv("HEIGHT", "1024"))
token_provider = get_bearer_token_provider(
    DefaultAzureCredential(),
    "https://cognitiveservices.azure.com/.default",
)
response = requests.post(
    f"{origin}/mai/v1/images/generations",
    headers={"Authorization": f"Bearer {token_provider()}"},
    json={"model": model, "prompt": prompt, "width": width, "height": height},
    timeout=180,
)

The full sample includes the matching Azure OpenAI route for gpt-image-* deployments and handles both base64 and URL image responses.

The full script

The attached text_to_image.py wraps this into a runnable script. Copy code/.env.example to code/.env and set your PROJECT_ENDPOINT and image deployment MODEL. Use an MAI deployment such as MAI-Image-2.6, or later switch the same setting to an Azure OpenAI image deployment such as gpt-image-1. If the model name contains mai-image, the script calls /mai/v1/images/generations; otherwise it calls /openai/v1/images/generations.

For short prompts, set PROMPT directly in .env. For longer multi-line prompts, especially prompts that contain quotation marks, save the text in a file such as prompt.txt and set PROMPT_FILE=prompt.txt.

OUTPUT_FILE is treated as the base filename. The script appends the deployment name and timestamp before the extension, for example generated_image_MAI-Image-2.6_20260924_160530.png.

python Code/text_to_image.py

Saved image to generated_image_MAI-Image-2.6_20260924_160530.png

Microsoft Learn resources

• Deploy and use MAI image models in Microsoft Foundry — deploy MAI image models and call the /mai/v1/images/generations API.
• MAI image API endpoints — reference the MAI image generations and edits endpoint shapes.
• Generate images with Azure OpenAI in Azure AI Foundry Models — quickstart for gpt-image-* image generation.
• Azure OpenAI image generation models — REST API setup and image-generation guidance for Azure OpenAI deployments.