Demo Image Masking

In this tutorial, you’ll learn how to use DALL·E 2’s image masking feature to edit an existing picture by defining a masked region and supplying a text prompt. This approach is perfect for seamlessly replacing objects, changing backgrounds, or adding new elements to your images.

Prerequisites

• Python 3.7+
• The openai Python package (pip install openai)
• An OpenAI API key

export OPENAI_API_KEY="your_api_key_here"

1. Install and Authenticate

Begin by installing the client library and initializing your API key in Python:

import openai
import os
from IPython.display import Image

# Load API key from environment
openai.api_key = os.getenv("OPENAI_API_KEY")

2. Prepare Your Images

You need three files:

1. Original image (with the object you want to replace)
2. Base image (same image after removing the object)
3. Mask image (a transparent PNG that marks the area to edit)

display(Image(filename='images/dog_table.png'))        # Original image with dog
display(Image(filename='images/table.png'))            # Image with dog removed
display(Image(filename='images/table-masked.png'))     # Transparent mask placeholder

3. Perform the Edit

Use the create_edit endpoint to fill the masked area based on your prompt:

response = openai.Image.create_edit(
    image=open('images/table.png', 'rb'),
    mask=open('images/table-masked.png', 'rb'),
    prompt='A cat sitting on a dining table chair waiting for food',
    n=1,
    size='512x512'
)
edited_url = response['data'][0]['url']
display(Image(url=edited_url))

4. Generate Multiple Variations

If you’d like several options, increase n and iterate through the results:

response = openai.Image.create_edit(
    image=open('images/table.png', 'rb'),
    mask=open('images/table-masked.png', 'rb'),
    prompt='A cat sitting on a dining table chair waiting for food',
    n=3,
    size='512x512'
)
for result in response['data']:
    display(Image(url=result['url']))

5. Parameter Reference

6. Generate Image Variations

Beyond masking, DALL·E 2 can also create variations of a single image without any prompt:

response = openai.Image.create_variation(
    image=open('images/table.png', 'rb'),
    n=4,
    size='512x512'
)
for img in response['data']:
    display(Image(url=img['url']))

For full details on both endpoints, see the OpenAI Image API Reference.