Transform Image with AI Prompt | Generated by AI

Home 2025.08

As of August 2025, OpenAI’s DALL-E models, particularly the newer “GPT Image” models (like gpt-image-1), have evolved to support more sophisticated image manipulation, including “image-to-image” transformations. While DALL-E 2 had a specific “variations” endpoint for this, the newer models integrate this functionality through a combination of image inputs and detailed text prompts.

Here’s how you would approach an “image-to-image” task like applying a Ghibli style using the OpenAI API:

The Core Concept: Image Edits with a Text Prompt

Unlike a simple text-to-image prompt, image-to-image with OpenAI’s API involves providing both an image and a text prompt that describes the desired changes. The key is to instruct the model on how to “reinterpret” the image in a new style.

Step-by-Step API Usage

The process typically involves a few key steps:

  1. Prepare your image: The image you want to transform needs to be in a supported format (e.g., PNG, JPEG) and meet the size requirements (often a square aspect ratio is best). You will need to provide this image to the API call.

  2. Craft a powerful prompt: This is the most crucial part. You’re not just saying “make this Ghibli style.” You need to describe the elements of the Ghibli style you want the model to apply. A good prompt will act as a guide for the AI, directing it on how to re-render the image.

    • Bad prompt: “Ghibli style”
    • Better prompt: “A magical forest scene in the style of Studio Ghibli. Use soft watercolor textures, a vibrant but gentle color palette with golden hour lighting, and add a whimsical, dreamlike atmosphere.”
    • Even better prompt: “Transform this portrait into a Studio Ghibli character, maintaining their essential features but styling them with the distinctive Ghibli aesthetics: slightly simplified facial details, expressive eyes, and a soft color palette. Use hand-painted textures and a nostalgic feel.”

  3. Make the API call: You’ll use the OpenAI API for image edits. The endpoint for this is typically part of the image generation API, but with specific parameters for image input. You’ll pass your image (often as a Base64 encoded string or a file ID if you’ve uploaded it to OpenAI’s server) and your detailed prompt.

    • Endpoint: The specific endpoint to use might be /v1/images/edits for DALL-E 2, but for newer models like GPT Image, it might be integrated into a single, more powerful /v1/chat/completions endpoint that handles multimodal inputs (both text and images). The documentation will specify the correct endpoint and how to structure your request.

    • Parameters:

      • model: Specify the model you want to use, such as dall-e-2 or a newer model like gpt-image-1.
      • image: The image data you’ve prepared.
      • prompt: The text description of the Ghibli style you want to apply.
      • n: The number of images to generate (often limited to 1 for newer models).
      • size: The desired output size (e.g., “1024x1024”).

  4. Handle the response: The API will return a JSON object that contains a URL to the newly generated image. You can then download and save this image.

Example Code (Conceptual Python)

While the exact code may change with API updates, here’s a conceptual example using the openai Python library:

import openai
import base64
from io import BytesIO
from PIL import Image

# Set up your OpenAI API key
# You should get this from your environment variables, not hardcode it
openai.api_key = os.getenv("OPENAI_API_KEY")

# Function to encode the image to base64
def encode_image(image_path):
    with open(image_path, "rb") as image_file:
        return base64.b64encode(image_file.read()).decode("utf-8")

# Path to the image you want to transform
image_path = "your_photo.png"
base64_image = encode_image(image_path)

# Your detailed Ghibli-style prompt
prompt_text = "Transform this image into a Studio Ghibli animation style. Apply soft watercolor textures, a vibrant but gentle color palette with golden hour lighting, and a whimsical, dreamlike atmosphere. Maintain the original composition of the image."

try:
    response = openai.images.generate(
        model="gpt-image-1", # or "dall-e-2" for the variations endpoint
        prompt=prompt_text,
        image_base64=base64_image, # This parameter might vary depending on the API version
        size="1024x1024",
        n=1,
    )

    image_url = response.data[0].url
    print(f"Generated Ghibli-style image: {image_url}")

except Exception as e:
    print(f"An error occurred: {e}")

Important Notes:

Back Donate