Meet DALL-E 2, the robot artist using AI to make dreams a reality

Have you ever wanted to paint a portrait of your cat in the style of Rembrandt Van Rijn but just didn’t have the time? Maybe you didn’t have the oil painting skills of Rembrandt, which are pretty hard to come by. No problem, DALL-E can do it for you. You can even start with an actual photo of your cat, so the portrait is as true to form as possible.

Koala astronaut holding a can of La Croix? DALL-E can paint that. Dinosaurs dressed like chocolatiers in Belgium? That, too. Tiny airplanes delivering toothpicks to patrons at a restaurant? You guessed it. DALL-E can make that a reality.

The AI ​​can instantly create these images in any artistic style or medium, including photography. The application uses natural language to create works of art, both an impressive and mindblowing feat.

DALL-E is an artificial neural network, or a combination of AI algorithms inspired by the biological network of nodes and neurons inside our own brains and bodies. The name was derived from a combination of Wall-E, the adorable Disney PIXAR robot, and Salvator Dali, a famous surrealist painter.




A bowl of soup that is a portal to another dimension as digital art © DALL-E 2

An astronaut riding a horse in a photorealistic style © DALL-E 2



An astronaut riding a horse in a photorealistic style © DALL-E 2

Robots vs Artists

Slow down, illustrators and digital artists. Before you chuck your Wacom tablets for greener pastures where artificial intelligence isn’t outdoing you, it’s important to mention that DALL-E does have flaws. It’s always important to remember that no matter how eerily close to human-AI may become, it can never be truly human.

To explain DALL-E’s shortcomings, let’s first unpack how it works.

Let’s Learn About Machine Learning

DALL-E is an incredibly intelligent machine that gathers images from the massive content well of the internet and sorts them according to their labels. Since the early days of the world wide web, users labeled images intentionally through meta text and alt text or unintentionally by engaging with them and sorting them ourselves (think Pinterest).

Over the years, AI engineering firms like OpenAI have been building machines to identify and short this content. Engineers also employ legions of web users to assist in labeling images by key identifiers. Over time and through lots of machine learning, DALL-E has built a massive library of specifically labeled imagery.

For example, you know without a doubt that if you google search Gwyneth Paltrow, her image will pop up. You know the same for ‘tennis’, and ‘aardvark’ google search queries, too. If you google searched ‘Gwyneth Paltrow playing tennis against an aardvark,’ it is incredibly unlikely that you will find an image that fits your vision. In a matter of nanoseconds, DALL-E gathers those three separate images and sifts through its library to find pictures of people playing tennis with each other. It then constructs an easily readable composition of a tennis match, then seamlessly swaps the players with Gwyneth Paltrow and an aardvark.

Teddy bears mixing sparkling chemicals as mad scientists as a 1990s Saturday morning cartoon © DALL-E 2



Teddy bears mixing sparkling chemicals as mad scientists as a 1990s Saturday morning cartoon © DALL-E 2

Art School for Robots

The most incredible thing about DALL-E is its ability to combine elements while still making an image that looks cohesive, readable, and creative. It can also utilize knowledge of different art styles, like creating a robot in the style of Picasso or making one person’s photo into seven different styles of a painted portrait. How is this possible?

Along with a vast library of content to draw from, DALL-E also uses algorithms to get smarter and smarter as time goes on. Let’s say 1 million people worldwide have visited museums with paintings by Vincent Van Gogh in them and posted a photo on their social media and wrote something in the caption about Van Gogh.

DALL-E now has 1 million examples of Van Gogh’s artistic style. It also has all the online libraries of high-resolution scans from museums and learning institutions. It can study every brushstroke, every variation in colour, and each way Van Gogh paints different things. When you ask DALL-E to paint a giraffe playing tiddly winks with manhole covers in the style of Vincent Van Gogh, the AI ​​will take those specific colors and brushstrokes combined with its knowledge of all the other visual elements and create an extraordinarily accurate rendition of the most bizarre Van Gogh painting ever.

Let’s use a more internet-centric example. Let’s say you want a photograph of yourself turned into an e-girl style portrait. DALL-E has likely sifted through Tumblr feeds and Twitter memes enough to know what an e-girl looks like and can output exactly what you want. Use Cases for DALL-E

Aside from creating imaginary paintings of animals doing human things, which is always delightful, DALL-E has many other potential applications that could change the visual media world. Currently, the AI ​​can only produce still images, but OpenAI’s next goal is to develop its video output, which would be even more monumental.

The Metaverse

The biggest use case for DALL-E is to grow the Metaverse. One of the biggest current issues with the Metaverse is that it’s growing faster than artists and developers can keep up with. Many people who enter the Metaverse now are less than impressed with the graphics and visual style of their surroundings. DALL-E can create detailed images of any space your mind can imagine, making the possibilities endless.

Video games

Another potential use for DALL-E is in video game graphics and world-building. For example, the development process for the game Cyberpunk 2077 took over nine years. Building virtual worlds out of nothing is no small task, and the capabilities of DALL-E could make this a much easier, shorter process with far more possibilities.

photo editing

Finally, DALL-E 2 makes the arduous process of photo editing way easier. In a demo, the AI ​​switches out a picture of a dog on a couch and a cat seamlessly. The bright side is cutting out the hours of work it takes to complete photo editing tasks. the downside? Our sense of reality through photographs seen online becomes more and more blurred. Think about the influencers who edit their photos to perfection, so people using social media apps think it’s possible to look like a photoshopped version of a human. Then, make that process faster and easier.

A bowl of soup that looks like a monster knitted out of wool © DALL-E 2



A bowl of soup that looks like a monster knitted out of wool © DALL-E 2

DALL E 2 can take an image and create different variations of it inspired by the original



DALL E 2 can take an image and create different variations of it inspired by the original

It’s Not Easy Being a Machine

DALL-E has three major shortcomings that should ease your mind if you’re an artist who is feeling threatened right now.

labeling errors

It’s easy to imagine that through the past few decades of large scale internet adoption, some images may have been labeled incorrectly. If enough people mistake a train for a monorail, you may ask DALL-E to paint a train, only to get a painting of an above-ground monorail instead.

A Lot of Machine Learning Left to Go

There is a possibility that certain topics or labels are such a niche that DALL-E may make a mistake in creating its artwork. It also may become confused for words with multiple meanings, unable to grasp context the way a human might. For example, you may ask DALL-E for a picture of two people on a date, and the AI ​​might output an image of two people on top of a giant piece of dried fruit.

There are also new topics and niche information that are too specific for DALL-E to grasp at the moment. If you want to create a painting of a very rare, endangered species of rainforest frog, DALL-E might not get it right. With time, that will get better and better as it improves its ability to sort and label content online.

An astronaut playing basketball with cats in space as a children's book illustration © DALL-E 2



An astronaut playing basketball with cats in space as a children’s book illustration © DALL-E 2

What is Art Without Humanity?

The most important difference between DALL-E and a human artist is its capacity to feel and respond to communication. Although DALL-E may be able to draw something similar to Tracy Emin’s artwork in style, a robot cannot experience it’s like for Tracy Emin to continue to create art after her cancer diagnosis. Therefore, the artwork doesn’t hold as much emotional power.

DALL-E could create a desolate cityscape similar to Max Ernst’s painting Europe After the Rain. Still, a machine could never know what it was like to endure the destruction of your home, family, and community as a European Jew during World War II.

In that way, DALL-E will never be able to compete with artists. Is art really art if there isn’t human experience or emotion behind it? DALL-E can most definitely become a tool for artists to express themselves in new ways. However, nothing could ever replace artists.

Leave a Comment