Cartoon generated by DALLE
Image generated using DALL·E.
Prompt: Funny cartoon of a designer using DALL·E and taking credit for the work


Visual Content Generation with DALL·E

DALL·E is a 12-billion parameter version of GPT-3 trained to generate images from text descriptions using pairs of text-images. DALL·E allows designers to generate different types of images that you have the right to reprint, sell, and merchandise. Where I have used AI generated images in my portfolio, I always make sure to call it out.

Below is the process I used to generate the photo on my projects page. This isn't meant to be an all inclusive review of DALL·E. Instead, I wanted to show my thought process and give an example of a few limitations. The below is in a short story format to show my thought process, the corresponding prompts, and a few limitations I ran into while trying to generate an image that looks like me working in a coffee shop - One of my favorite settings to do work. Enjoy!

first image generated

When I asked for a web developer the first image DALL·E generated was a man in a coffee shop and I wanted the web developer to look more like me. Plus the code from the computer screen is also on the back of his t-shirt. Let's fix this!

second image generated

This is looking better but the code is still out of place (you can see it on the back of the computer). So I asked DALL·E to change the angle of the picture.

first image generated

This looks like an idealistic setting. However it reminded me more of Europe, and less of Philadelphia, which is close to where I am based. I asked DALL·E to revise the image further.

first image generated

Excellent! I really liked this image but I wanted something less idealistic, and more realistic. If you notice there's a bunch of steaming cups and no people to drink them although there's lots of people outside of the window.

first image generated

I had to iterate a little to get this image. I asked for a Norman Rockwell style image and ChatGPT can't reproduce a specific style of an artist. Instead it prompted me back and supplied key words I could use to generate the style I was looking for. Very helpful DALL·E!

This is looking great again, except for an important detail - The web developer has no computer now! So asked DALL·E to add it back in.

sixth image generated

This photo would probably work if you aren't familiar with Philly. However you'll notice that the William Penn tower in the back has been duplicated. There's only one William Penn tower in Philly, not two. Instead of removing it, I changed the background to another iconic Philly landmark, Love Park.

seventh image generated

Overall, DALL·E is a very powerful image generation tool that can level up your digital content. Plus you own the content you generate. A few other limitations I noticed:

  • DALL·E can't generate very realistic photos. For example, I tried generating stock photos of employees and they always looked like cartoons even when I prompted for, "an ultra-photorealistic image" or "--style raw".
  • You need a human to pay attention to the details because DALL·E generated images that often had details that didn't make sense or lacked the context a human has.
  • Content limitations exist, for example I could create images in the style of Van Gogh's paintings, however for more recent artists like Andy Warhol or Norman Rockwell, you have to describe the characteristics of their art to generate the desired style.

Thanks for reading!


That's all folks!
Image generated by DALL·E.