Last updated February 09, 2026

The Stability AI Stable Image Ultra (stability-image-ultra) model allows you to generate high-quality, detailed images from descriptive text prompts. This guide shows how to access the v1-images-generations API using Python.

Prerequisites

1. Attach an inference addon to an app of yours:

   # If you don't have an app yet, you can create one with:
   heroku create $APP_NAME # specify the name you want for your app (or skip this step to use an existing app you have)
   
   # Create and attach one of our chat models to your app, $APP_NAME:
   heroku addons:create heroku-inference:standard -a $APP_NAME --as INFERENCE
   

2. Install the necessary requests package:

   pip install requests
   

Python Example Code

import requests
import json
import os
import base64
import subprocess
import sys

# Global variables for API endpoint, authorization key, and model ID from Heroku config variables
ENV_VARS = {
    "DIFFUSION_URL": None,
    "DIFFUSION_KEY": None
}

# Assert the existence of required environment variables, with helpful messages if they're missing.
for env_var in ENV_VARS.keys():
    value = os.environ.get(env_var)
    assert value is not None, (
        f"Environment variable '{env_var}' is missing. Set it using:\n"
        f"export {env_var}=$(heroku config:get -a $APP_NAME {env_var})"
    )
    ENV_VARS[env_var] = value


def open_image_file(filename):
    """
    Opens an image file using the default image viewer based on the operating system.
    """
    try:
        if sys.platform == "darwin":  # macOS
            subprocess.run(["open", filename], check=True)
        elif sys.platform == "win32":  # Windows
            subprocess.run(["start", filename], shell=True, check=True)
        elif sys.platform == "linux":  # Linux
            subprocess.run(["xdg-open", filename], check=True)
        else:
            print("Automatic image opening is not supported on this platform.")
    except Exception as e:
        print(f"Failed to open image: {e}")


def parse_image_output(response, payload, open_image=False, filename=None):
    if response.status_code == 200:
        result = response.json()
        if payload.get("response_format", "url") == "base64":
            # Generate a default filename if none provided, based on the prompt
            if not filename:
                filename = payload["prompt"][:20].replace(" ", "_").lower() + ".png"
            with open(filename, "wb") as f:
                f.write(base64.b64decode(result["data"][0]["b64_json"]))
            print(f"Image saved as {filename}")

            # Open the image if flag is set
            if open_image:
                open_image_file(filename)
        else:
            print("Download the image from:", result["data"][0]["url"])
    else:
        print(f"Request failed: {response.status_code}, {response.text}")


def generate_image(payload, open_image=False, filename=None):
    """
    Generates an image using the Stability AI Stable Image Ultra model and saves it to a file.

    Parameters:
        - payload (dict): dictionary containing parameters for the image generation
        - open_image (bool): flag to open the image automatically after generation
        - filename (str): name of the saved file (defaults to a truncated version of the prompt)

    Returns:
        - saves the generated image to a file and optionally opens it
    """
    # Set headers using the global API key
    HEADERS = {
        "Authorization": f"Bearer {ENV_VARS['DIFFUSION_KEY']}",
        "Content-Type": "application/json"
    }
    endpoint_url =  ENV_VARS['DIFFUSION_URL'] + "/v1/images/generations"
    response = requests.post(endpoint_url, headers=HEADERS, data=json.dumps(payload))

    parse_image_output(response=response, payload=payload, open_image=open_image, filename=filename)


# Example payload
payload = {
    "model": "stable-image-ultra",
    "prompt": "A surreal landscape with glowing mushrooms under a night sky.",
    "aspect_ratio": "16:9",
    "output_format": "png",
    "seed": 123,
    "negative_prompt": "crowded, noisy, chaotic"
}

# Generate the image with the given payload
generate_image(payload, open_image=True)