“When you ask ChatGPT's DALL-E to make a picture more of something over and over again, things get weird.
AI art generators are in a weird place. They can attempt to make just about anything you can think of, from a dog skateboarding in outer space, to a cup of coffee floating in the ocean. Putting the ethics of AI art aside, some of these creations do not hit the mark on the first go around, and you need to prompt the AI bot with changes to tweak the final results to your liking.
But what if your end goal isn't to produce a quality piece of AI art? What if your goal is to make something wild.
That's what the "make it more" trend is all about. ChatGPT users are asking DALL-E to generate an image, then once that image pops out, they ask the bot to make it more of something. In this example from Justine Moore, DALL-E was prompted to create a bowl of ramen. After that initial prompt, Moore asked it to make it spicier. It followed suit, mostly by adding a lot of peppers to the mix. She again asked DALL-E to make it spicier. It complied by setting the bowl on fire in what appears to be Pepper Hell™. By the end of the exercise, the bowl of ramen was shooting fire beams into outer space, a truly spicy bowl of noodles.
There are plenty of examples of this trend online to peruse for your pleasure, from Mashable editor Stan Schroeder's gigantic water bottle experiment, to this body builder getting more and more buff. If you want to try the trend for yourself, however, you should be aware of some constraints.
How to use the "make it more" trend with ChatGPT
First of all, DALL-E, like other elements of GPT-4, has a limit to the amount of prompts you can issue at any one time. OpenAI isn't super clear when you're about to hit your limit, but just be careful not to get too carried away with your experiments, or else you'll need to wait a few hours to try again.
Second, DALL-E is finicky with this type of request. I'm not sure if this is something OpenAI adjusted since the trend picked up steam, but I've had trouble getting DALL-E to cooperate with making a piece of art more of something. I tried two different scenarios in particular. First, I asked the bot to generate an image of a dog running through a field. It did. I then asked it to make the dog faster. It complied. I asked it to make the dog faster again, but it rejected me, letting me know that it already made a dog that was fast, and didn't feel the need to make it go faster.
Credit: Jake Peterson
I tried the dog trick again, asking it first to create the fastest dog in the world, then asking it to make the dog even faster. DALL-E rejected me again, saying it had already made the fastest dog in the world. Silly me.
I had more luck asking the bot to generate a cup of coffee, then asking it to keep making the coffee hotter. At first, it tapped out after a couple of iterations, but I was finally able to get the bot to generate about five progressively "hotter" cups of coffee. By the time it told me that it couldn't represent heat any differently, the cup looked like it was undergoing the Trinity Test:
Credit: Jake Peterson
I encourage you to try the trend out for yourself in the AI generator of your choice. Just remember: Start small (e.g. "generate a cup of coffee"), then ask it to change it in a simple way ("make the coffee hotter").