OpenAI, ChatGPT, DALL·E 2, create image from text prompt

Published: Update on: Classify at:
Words:326 Read:≈ 2min Views:

This post is describe how to generate image from text prompt.

How to use OpenAI

  1. Go to the website openai.com and create user profile.
  2. Generate new OPENAI_API_KEY.
  3. Use public API

DALL·E 2

DALL·E 2 is a revolutionary artificial intelligence system that can generate realistic images and artwork based on descriptions provided in natural language.

Install OpenAI CLI

1
2

pip install openai
export OPENAI_API_KEY=my-secret

OpenAI CLI create image

1

openai api image.create -p "Artificial Intelligence, future, cyberpunk, with hex color 33FFE9 and color pink" -n 4

Docs api image.create

OpenAI Python script create image

Some fun Python code to work with Rick and Morty and OpenAI Image API.

 1import shutil
 2import pathlib
 3import openai
 4import requests  # pip install requests
 5from PIL import Image
 6
 7HOME = str(pathlib.Path.home())
 8openai.api_key = "MY_SECRET_KEY"
 9
10def save_image(url, name):
11    r = requests.get(url, stream=True)
12    if r.status_code == 200:
13        with open(f'{HOME}/Downloads/rick/{name}.png', 'wb') as f:
14            r.raw.decode_content = True
15            shutil.copyfileobj(r.raw, f)
16
17
18image_urls = []
19prompts = [
20    "Rick and Morty stuck in a time loop",
21    "Rick and Morty sitting on a couch, looking frustrated",
22]
23for prompt in prompts:
24    response = openai.Image.create(
25      prompt=prompt,
26      n=4,
27      size="1024x1024"
28    )
29    for data in response['data']:
30        image_urls.append(data['url'])
31
32for i, url in enumerate(image_urls):
33    save_image(url, i)
Rick and Morty stuck in a time loop Rick and Morty stuck in a time loop Rick and Morty stuck in a time loop Rick and Morty stuck in a time loop Rick and Morty sitting on a couch, looking frustrated Rick and Morty sitting on a couch, looking frustrated Rick and Morty sitting on a couch, looking frustrated Rick and Morty sitting on a couch, looking frustrated

API openai.Image.create_edit has some limitation, so we need to compress input image. Input images:

1foo = Image.open(f"{HOME}/Downloads/rick/image_edit.png")
2foo = foo.resize((1080,1080),Image.LANCZOS)
3foo.save(f"{HOME}/Downloads/rick/image_edit2.png", optimize=True, quality=75)
4
5foo1 = Image.open(f"{HOME}/Downloads/rick/image_mask.png")
6foo1 = foo1.resize((1080,1080),Image.LANCZOS)
7foo1.save(f"{HOME}/Downloads/rick/image_mask2.png", optimize=True, quality=75)

Size optimized images:

and then we can make API call

 1image_urls2 = []
 2response = openai.Image.create_edit(
 3  image=open(f"{HOME}/Downloads/rick/image_edit2.png", "rb"),
 4  mask=open(f"{HOME}/Downloads/rick/image_mask2.png", "rb"),
 5  prompt="Rick and Morty stuck in a time loop",
 6  n=4,
 7)
 8for data in response['data']:
 9    image_urls2.append(data['url'])
10
11for i, url in enumerate(image_urls2):
12    save_image(url, i)

Result is not so great but if you spend some time to create good prompt, you will get better one.


Links