Meet DALL-E, the AI sentence-image generator

Last Updated on: 9th August 2022, 11:27 pm

If you’ve been on the internet in the past few days, you’ve probably stumbled across a collection of images showing some ridiculous scenarios, such as camera footage of Jesus Christ stealing a bicycle, and Fisher-Price versions of not-so-kid-friendly objects such as guns or random characters committing crimes. 

These are all AI-generated images by DALL-E, a powerful transformer language model by OpenAI written in Python

By imputing any prompt, it starts to generate images containing the exact description contained in it. Or in simpler terms, you type a sentence and it gives you images based on what’s written, no matter how absurd the sentence could be.

What is DALL-E?

DALL-E is described as a “brother” to GPT-3, another language model focused on creating texts, but it has a focus on creating images. DALL-E is named after both the Disney robot Wall-E and the surrealist artist Salvador Dali. And it couldn’t be better described as an AI putting the unimaginable on a picture.

Dall-E is trained to create images from any text descriptions, as it’s described on its initial page with pretty simple but interesting prompts. Any concept can be combined, creating some interesting images and situations.

But what happens when the whole internet discovers this AI and becomes obsessed with it? Well, this happens:

The old saying about how humor would be randomly generated in the future wasn’t wrong after all. Twitter pages and subreddits dedicated to weird and funny DALL-E generations are becoming more and more popular as the AI somehow indirectly embraces the 21st-century sense of humor.

READ MORE: Xbox gamers aren’t happy with Final Fantasy 7 Rebirth announcement

How can I get to Dall-E?

The quickest way is through DALL-E mini, an online version of the original application. While it’s not as perfect as the original and has some traffic problems, it’s quite effective for a quick test of the AI.

The original does have its source code available on Github, but not a perfect version, and you could find trouble by trying to run it yourself. 

DALL-E 2 is an enhanced version, and you can get access to it by entering the waitlist and patiently waiting for them to get you an invite. 

Are there limits?

As long as you don’t generate inappropriate images that go against the platforms’ ToS, you can virtually create anything you want. The AI was never meant to become a meme-maker, and it contains huge potential for machine learning, creative creation, and even education at some point.

You may face some problems as simpler inputs may not give what you wanted, but over-detailed sentences will not always work either, so finding balance is the key. 

The site could also occasionally generate offensive images, as stated by the DALL-E site:

“While the capabilities of image generation models are impressive, they may also reinforce or exacerbate societal biases. While the extent and nature of the biases of the DALL·E mini model have yet to be fully documented, given the fact that the model was trained on unfiltered data from the Internet, it may generate images that contain stereotypes against minority groups. Work to analyze the nature and extent of these limitations is ongoing, and will be documented in more detail in the DALL·E mini model card.”

As the application has become more popular recently, OpenAI may decide to share more information on the exact process that it was created in order to help us understand exactly how it does what it does. 

But if you don’t care about that, you can just have fun with Shrek chilling with Obama.