Why the DALL-E neural network won’t take my job as an illustrator

Finally I was able to test DALL-E. I can assure you that it is impossible to generate professional and usable images from a text description. Here are my attempts to be creative with the help of a neural network.

I have been a professional illustrator and storyboard artist for ten years, working primarily in advertising and film. I was really scared when I started reading all those articles about DALL-E and Midjourney that said AI was going to take my job as an illustrator. But when I got acquainted with how neural networks work, I became more skeptical about this. So I signed up for the queue and finally got a chance to use them myself to see if I could replicate what I do for a living using artificial intelligence.

I just tried to recreate professional images that needed to be done for a client, especially a simple storyboard frame for a Citroen car company. The choice, quite by accident, fell on a frame that was made for an Instagram ad (that’s why it’s vertical). This is what the original looks like:

As you can see, the image is very easy to describe. This is a man and a woman in a Citroen Ami, front view. I don’t even mind that the car is blue and the characters are white. I just wanted to see if I could get something similar from DALL-E. The results were better than I expected, but far from being helpful to my client. Here are some of the results I got with different descriptions:

The AI ​​didn’t know the exact model of the car (probably because it’s new). The characters are ok, but not particularly realistic. It was almost impossible to get identical images. I also forgot to mention a lot of details that seemed intuitive to me, such as the fact that the characters are smiling, looking ahead, the woman is driving, the driver and passengers are wearing seat belts, and so on.

So, all this to the fact that it is very, very difficult to describe the image in words. There are so many things that an illustrator knows instinctively, but the machine needs to be taught. Even if it were possible, it would take a very long time to get a coherent and satisfying description, and a lot more time to run tests and get the right result. It took me 10 to 15 minutes to generate these images (besides thinking about the description, the image generation itself took several minutes). None of them are suitable for professional presentation. I spent about half an hour to draw everything myself, including a sketch, to check with the client for the correct angle and composition, as well as edits based on his feedback.

Now imagine having to go through this process over and over again to finally get a cohesive storyboard of 12 images like this, with the same characters and cars from different angles:

I doubt very much that such software, even greatly improved, will really help my clients professionally. Perhaps this will allow them to create sketches faster, but they are unlikely to want to save money and get something like that.

They need quality, accurate and professional sketches that represent their vision. They run very expensive ad campaigns (ranging from a few hundred thousand dollars to millions when it comes to TV), so I don’t think they would save a few hundred (or thousand) dollars by replacing their artist with artificial intelligence. Clients want everything to be professional and to the point so they can communicate the idea to their boss, clients, director, camera crew, special effects department, and so on.

A storyboard is a professional tool that allows you to create a holistic vision of a project. It may not be inaccurate or approximate. And I’m already silent about the need to depict the same characters with pronounced emotions, as in these sketches that I did for Brico Depot.

It’s reminiscent of a time when cameras got cheap and everyone had a good camera on their phones, which made everyone think it was going to kill professional photographers. But it was not so.

People still hire professional photographers for their weddings (my sister recently did) because pictures look so much better when they’re taken by a professional. And advertising agencies and fashion brands still pay big bucks to professional photographers to shoot their campaigns. I doubt that any serious company will simply say: “Let’s just buy a good camera and do everything ourselves, it will be cheaper than hiring a real photographer.” A tool is just a tool. You still need someone who specializes in its use and has a good vision.

I actually took a few photography lessons during my studies. I had a very good camera and I can assure you that it is very, very difficult to take good pictures. I confess that I am a bad photographer. It’s just not mine. If I needed professional photos, I would hire a real photographer. A person is a creator not because he has a good tool.

The same could be said about artists. When photography was invented, people mistakenly assumed that paintings would no longer be in demand. When digital art was created, traditional art continued to evolve. They are just different forms of creativity with their own specialists.

Speaking of art forms, how is DALL-E doing with art drawings? I tried to recreate my own illustration, because in this case there is less reason to be precise (since it’s not for work). Just wanted to know if DALL-E would perform better in art than in a professional storyboard. So here’s the image (I’m a big fan of Vaporwave and Lofi Aesthetic, I make a lot of these 80s style pics for my Instagram and then animate them).

And here are the images that DALL-E took with my description of the picture:

I don’t even know what to think. It’s just not what I had in mind. Of course, it happened, because it is almost impossible to explain the image from the head in words. Also, it’s obviously not my style, the image lacks personality. But, frankly, for AI, the result is impressive. It will probably get even better with time. However, do you understand what I’m talking about when I say that it will be many more years before people can create interesting works using these tools?

Then I tried something else. Out of curiosity, I uploaded the same image and asked DALL-E to create alternate versions of it. Here are the results:

Yeah… Take a closer look at one of them. I think you will see the problem, haha.

You may ask, what if AI gets much better in the future? As I said, even chess has survived the success of artificial intelligence. When Deep Blue beat Garry Kasparov in 1997, everyone assumed that chess was going to die. However, today more and more people are learning to play them. There are so many chess channels on YouTube and Twitch that have millions of subscribers. And world champions like Magnus Carlsen are still making millions of dollars.

At the same time, no player can compete with AI. In fact, they use a neural network to learn how to better understand the game.

So why isn’t chess dead? Because in sports, as in art, people are more interested in human abilities, even if they are nothing compared to a machine or animals. Almost no one watches chess matches between computers, it’s boring.

In addition, today with a pistol you can kill the most powerful fighter on the planet in a second. An experienced martial artist who has spent 30 years perfecting his skills can be defeated by a child who finds a weapon. We would assume martial arts are dead as they are useless due to new technology. But we still love to watch sports fights where opponents fight with their fists. Because what matters is not the end result (getting stronger or defeating the opponent), but the human superiority that people admire. Killing someone with a gun is not impressive. It’s much more interesting to watch someone do it with their own hands against a similarly trained fighter.

A car can lift a house, but people still love to compete in and watch strongmen. A car or even a cheetah will easily overtake Usain Bolt, but he still impresses us as being the fastest man on the planet. We don’t care if some Joe is faster on his bike. We are impressed that a person can run faster than anyone else.

So, even if AI becomes better at creating illustrations than humans (I’m not saying it’s impossible, just skeptical), I still believe that my field of activity will survive.

The coolest thing will probably be to no longer see the final results, but to look at people in the process of drawing. Some automatic pianos can play anything, but people still enjoy watching a real person play. It is much more pleasant to watch how people do or create something, and not machines.

Therefore, all people who talk about the death of illustration and animation should know that art cannot die. It’s an idea, and ideas don’t die. Moreover, this idea has been central to human life since the beginning of time. Ever since the guy in the cave figured out 50,000 years ago that he could leave a handprint on a wall with dirt or blood or whatever. Or that he can sing and dance and tell stories…

AI is a tool. Art is characterized not by the tool, but by the will to art, which the machine does not have. The tool can transform it, improve it, expand it, but not erase it. I am very interested in what AI will bring to future artists. I don’t care if they steal our jobs. Okay, maybe deep down it’s a bit of a worry, but at least it won’t happen in the near future, I’m sure 🙂

Have a good day!