最終更新日 2025年01月24日(金)

Stability AI Stable Image Ultra​ (stability-image-ultra​) モデルを使用すると、説明的なテキストプロンプトから高品質かつ精細な画像を生成できます。このガイドでは、Python を使用して v1-images-generations​ API にアクセスする方法について説明します。

前提条件

1. まだインストールされていない場合は、Heroku CLI​ をインストールします。次に、Heroku AI プラグインをインストールします。

   heroku plugins:install @heroku/plugin-ai
   

2. 画像モデルをアプリにアタッチします。

   # 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 the image model to your app, $APP_NAME.
   heroku ai:models:create -a $APP_NAME stability-image-ultra --as DIFFUSION
   

3. 必要な requests​ パッケージをインストールします。

   pip install requests
   

Python のサンプルコード

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,
    "DIFFUSION_MODEL_ID": 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": ENV_VARS["DIFFUSION_MODEL_ID"],
    "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)