Art with the help of Dall-e Artificial Intelligence

My brief artistic side

In mid-2014, during my technical course, I had my first contact with design tools like Adobe Photoshop, Dreamweaver, and others. I even created some cool images, and if I hadn't pursued programming, I believe I would have focused more on graphic design and illustration.

In 2021, I decided to learn how to draw and followed the book You Can Draw in 30 Days: The Fun, Easy Way to Learn to Draw in One Month or Less (non-affiliate link) and really learned a lot from the basics. I realized that it's not as difficult as we think and that with a little practice and technique, we can achieve really cool results.

But beyond that, I never went much further in graphic illustrations and design. Of course, as a FullStack developer, I have to know the minimum of what is acceptable or not on a website, for example. But that's it, nothing worthy of a frame.

Generative AI tools

For a while, I've been using GitHub Copilot, so I'm already familiar with generative Artificial Intelligence (AI), even if superficially.

Recently, I learned about Dall-e (a mix of Salvador Dali and Wall-e), a tool that uses deep learning to generate images that correspond to an initial text. After doing some research and seeing some results, I was eager to test it out and applied to participate in the early preview.

After a month in the queue, I finally received my access to test it out. I got 15 tokens, and these tokens are spent on each generation made.

Generating images with prompts

To generate, just access here, and in the text box, insert the prompt.

For each prompt, 1 token is spent, and 4 images are generated. After that, we can try to generate variations of one of the images or insert another text.

The first prompts I did were a sadness. But I was learning. I tried to make extravagant images like:

"A water bottle fused with a zeppelin chasing an F1 car"

"A cat driving an F1 car in a cyberpunk track chased by a dog driving a monocycle with a monocle"

The truth is that I didn't even know what to expect, and I'm still trying to understand what happened in this cat image.

Improving prompts

Well, I realized that the problem must be in the prompts; since it is a generative AI, if I improve and be a little more descriptive, I should have better results.

That's when I came up with the idea of ​​generating a cover for my Spotify playlist.

"A human with headphones programming the future in a retro wave style"

It's already much better.

It's worth noting that Dall-e is not very good at generating images that contain text, so it's not even worth trying. Here's an example of logos I tried:

“minimalist logo, hexagonal, space, coding, purple”

“minimalist logo, shapes, space, planets, coding, purple”

I didn't even ask for text in the logo, but you can tell it's not good at it. But it's nice to have an idea.

A bright future

The trend is for images to get better and better as technology advances, and I'm impressed with what is already possible to generate nowadays.

There are debates emerging about whether this is art, whether it's not, how it was trained to learn what is what, and so on, but honestly, I believe they are tools. A designer could easily generate images as assets or to avoid that famous "creative block" and, with that, have some idea.

As a dev, I already use GitHub Copilot to help me with the more tedious parts of my day-to-day.

Imagine an AI that generates videos; there is likely already some company out there doing it.