DALL-E 3: A ChatGPT Experiment in Image Modification—Witness Its Ability in Action

I have recently been experimenting with using DALL-E 3 inside of ChatGPT Plus. I have been doing this as part of my work, not because I have an unhealthy addiction to seeing my mental descriptions come to life on the screen. I assure you, I can stop at any time. We have reached a point in the future where an AI says the equivalent of, “I’m sorry Dave. I’m afraid I can’t do that.” But today, I found a new toy. DALL-E 3 inside of ChatGPT can read and modify images, to a certain extent. However, it does have a few quirks. Let’s start this story at the beginning…

I have been using Midjourney to customize images for some time, but the process is quite convoluted. On the other hand, in ChatGPT Plus, all I have to do is click on the paperclip icon and upload the image. It’s much easier to use and also quite fun. But does it work well? To test it out, I tried modifying three images: a picture of my car, a picture of myself, and the ZDNET logo. Here are the results.

When I uploaded the image of my car, I asked DALL-E 3 to put the car in a city, and it successfully recreated an image of the car in a city scene. After that, I requested DALL-E to give it a steampunk look, which it did. However, I encountered issues with DALL-E where it failed every two or three requests, so I had to work around that.

Next, I uploaded the steampunk version of my car and requested DALL-E to make the car fly, resulting in a Chitty Chitty Bang Bang-style vehicle. I then tried to place Santa in the driver’s seat, but this request was denied.

When I made a combined request for placing the car in a city, turning it into steampunk, making it fly, and adding Santa to the driver’s seat, the result was a nice picture, but it didn’t fit the Victorian aesthetic of the previous images.

DALL-E essentially rewrites every prompt into a more detailed instruction. When I asked to put a picture of myself in an office setting, the request was denied. I then uploaded the ZDNET logo and tried to place it on a jet and a building, but the results were not what I expected.

In conclusion, after experimenting with DALL-E’s image modification feature, it is clear that there are some limitations. Have you tried uploading images to DALL-E? Let me know in the comments below.