Here’s how OpenAI’s DALL·E image generator works

If you are a graphic designer, you might want to stop reading now. The company OpenAI has just launched an artificial intelligence model (DALL·E) capable of generating images from texts. Let us explain how it works.

What is DALL·E?

DALL·E breaks new ground as the first creative artificial intelligence model capable of generating images from texts. In other words, it combines both an understanding of natural language with the generation of realistic images.

But DALL·E’s artistic talents don’t end with a simple snail drawing. If you wanted, it could generate multiple harp-shaped snail illustrations, or something more bizarre than you could even imagine. Just provide a description and it will generate a number of alternatives.

It can generate images of anything you can think of. And when we say anything, we mean absolutely anything! This model can:

• Transform existing images
• Create anthropomorphic animals and objects
• Combine concepts which are seemingly unrelated
• Represent text

Where does the name ‘DALL·E’ come from?

The name DALL·E is a combination of the surrealist painter Salvador Dalí and WALL·E, the PIXAR robot.

Why does it represent such a breakthrough?

Until now, artificial intelligence was able to generate images based on ones it had seen before. For example, using a dog database, we could train an AI model to draw a picture of a dog for us. Similarly, we could use a plant database and train the model to generate an image of a plant.

But DALL·E goes much further.

An avocado-shaped chair?

On its blog, OpenAI explains how DALL·E works through some very diverse – and quite surreal – examples. Here is a sample of them:

• An illustration of a baby daikon radish in a tutu walking a dog. Could the human mind come up with something like that? We’re not sure, but DALL·E certainly could. And this is the result.

• An armchair in the shape of an avocado. This image has spread like wildfire on social media. Many media outlets have used the image as a headline due to its popularity.

• The exact same cat on the top as the sketch on the bottom.

• A store sign with the ‘OpenAI’ on it. Very interesting for business owners.

Let’s not forget that these results have been both imagined and created by artificial intelligence. It has been worth our time going in to play around with the examples published by OpenAI, although it’s still not possible to enter free text, we can experiment with some of the terms.

How do you achieve all this?

It’s no surprise that these developments are the work of OpenAI, an artificial intelligence company owned by Elon Musk which aims to promote AI for the benefit of society.

We have already seen some of the advancements from this San Francisco-based company in the form of the GPT-2 and GPT-3 models. 

As such, we can assume that these were not easy to achieve without some significant backing behind them. Here are some figures on the DALL·E model:

• 12 million parameters
• 1,280 tokens (256 for text and 1024 for images)

The future of illustration

OpenAI has not yet published an article detailing the results, but for now we have this article. Unlike what they gave us with GPT-3 in 2020, this model comes with no public or private beta.

Although the company does concede some limitations and certain human intervention during training, OpenAI still serves up some surprising results. Many are already talking of a ‘revolution’ in the world of illustration and photography. While, at least in the short term, this by no means spells an end to graphic design, what it does mean is that it will have to adapt.

Today it’s images, but in the future it could be videos. We should proceed with caution, as such progress is a double-edged sword. However, its arrival seems inevitable and OpenAI is determined to prove it to us.

What’s next? Narrativa will keep you informed.