Table of Contents
I’ve been working with Dall-E for a few months now, and I think the time has come to give my review.
First off, I’d like to say the wonderful world of AI for all sorts of purposes is very fun. It’s a great time saver, and I think it will both revolutionize the way many of us work… but will also damage most of us in many, many ways as well. I think people are already becoming just a little less creative. I mean, why bother creating when you could have a machine do it for you just as well, much quicker, and for little to no money?
Technology: To Embrace or Not to Embrace
It’s really an age old question. And in the end we are forced to choose to embrace technology, or just watch as the world advances around us.
Humans need movement. When technology gave us lives void of movement, we needed to choose to find other ways to move… or become unhealthy and slovenly.
So too, humans need to be creative. So if our computers are going to do all of our creative work for us, we will need to find alternative outlets for our creativity. Or the world will become routine and boring, which would be awful. But I have faith we will find ways to enhance our lives without damaging the essence of humanity. Perhaps we’ll one day look back and we’ll be so much farther than we could have ever expected.
So what is Dall-E? It’s Chat GPT’s text-to-image software. You put in a prompt, and shortly after it either spits back an image… or it tells you your request is in violation of its policies. So if I say “Create a cartoon of a hamster eating a muffin while riding on a scooter, wearing a top hat, with dozens of dancing caterpillars in the background”, I might get something like this.
Dall-E does some things amazingly well… and some things far less so.
How to Use Dall-E
Here’s how I’ve found to use it correctly: Stick to animations. It can create incredible, detailed, sometimes shockingly high-quality animations.
Be meticulous about forming intelligent prompts. This is true, in my experience, of a lot of AI. If the program can misinterpret what you’re looking for, it often does. For some of my print-on-demand work, I use Chat-GPT to devise tags. A list of 15. If I just told someone to make the list, they could probably figure out how to format it intelligently. Chat GPT needs to be told that words needs to be separated, individual tags should have commas in between them, and after the comma there should be a space.
Is doing this fun? Not particularly. But you only have to do it once! After that you have a usable template that’ll work every time.
So if you’ve crafted a beautiful prompt and you want an animation, you’re in luck. If that’s what you’re looking for, you can get some amazing results from Dall-E.
Dall-E’s Imperfections
There are four things, however, where I think Dall-E is really not so good. We’re in the first generation of this technology. I have no doubt there will be extreme improvements. But in the meantime, if you want to try this out, be wary of the following four items.
First, Dall-E is terrible about letters and spelling. For whatever reason, 80-90% of my requests with words turn up everything from misspellings to absolute gibberish. Mind you, this is even when I put the word into the prompt explicitly. Dall-E is very smart… but borderline illiterate.
So if your goal is to make something like custom business cards or flyers for an event, I suggest you use Dall-E to generate the images, then use those images in a different software.
Mistakes
I’ve already mentioned that Dall-E struggles with making things that seem like authentic, realistic photos. But even though the cartoons are excellent, they’re often riddled with errors. An extra finger here, a detached tail there (notice anything funny about the zebra?). Sometimes the mistakes are very obvious. But other times you need to be extremely careful. You could end up publishing something or using an image promotionally not realizing the glaring error hidden within.
So the important thing is to make sure to carefully check each and every image you generate.
Restrictions
Possibly my biggest issue with Dall-E surrounds its uber-restrictive policies. I constantly have to generate and regenerate prompts to get something close to what I was looking for. There are so many rules, many of which are highly subjective. I was constantly walking into walls trying to get something even remotely resembling what I was looking for.
It brings me back to my teaching days when I had students who were trying to make PowerPoints about certain biblical stories. Every time someone did a Google search for “sword”, they got locked out of the system. Why? Apparently it’s a term used in some gay pornography. And because of that relatively obscure use of the word, dozens of children will have their education interfered with unnecessarily.
It’s important to always look deep within policies. Yes, I understand that you want to avoid violence and “sensitive” topics; however, maybe be a little less restrictive on what defines those terms. If not, we’re heading in a direction whereby soon you’ll get an error message for the prompt “Create a cartoon bird”. Somebody, somewhere will be able to find what’s offensive there!
To Dall-E or not to Dall-E
Finally, the program has a problem with consistency. So, for example, if you wanted to create a character and multiple images using the same likeness, either this is impossible or I haven’t yet figured out how to do it. This would be rough if you wanted to create a theme for a particular marketing campaign or wanted to illustrate a children’s book.
So, in my final assessment: If you are looking for individual animated images, and you have the patience to craft the perfect prompt and possibly do it a few times until you like the results, Dall-E is for you and can produce marvelous things. But it is filled with limitations. I have no doubt the program will continue to get better and better. And it very well might be the best that’s out there at the moment. In the meantime, enjoy what we’ve got. Learn to master what’s in front of you. And embrace the world of AI. It’s not going anywhere!
