It is a never-ending human desire to develop art. From being the pastime of people to helping to get through hard times to being the only remainders of our ancestors, art, especially visual art, has been the one thing that has bound people, the only thing everyone agrees on. And it is quite a prejudice that machines cannot create art that is meaningful. DALL E 2 is the new AI technology that is destroying this notion.

Shortly, DALL E 2 is a generative language model that creates realistic images from existing ones. 

Expanding on it, it is a neural network algorithm developed by the research laboratory Open AI in 2022. It is the successor to DALL E, a generative language model that uses natural language to create realistic images. It uses existing images on the internet to source and manufacture the required images. DALL E 2 is smaller than DALL E, with about 3.5B parameters while the latter had 12B, making DALL E 2 one of the much smaller models. While there are many other generative models (like Artbreeder), what makes DALL E 2 unique is its lifelike images that have a natural aesthetic quality.

Its name comes from a combination of the words ‘Dali’ (Salvador Dali, the Spanish Surrealist painter) and ‘WALL E’ (the Pixar movie). Personally, it represents the harmonious co-existence of art created by humans and the art created by technology.

Right now, only selected people have been given access to DALL E 2 for testing and experimenting purposes. Other people interested in using it have been added to a waiting list which is only increasing every minute. While 1000 new people are being given access to it every week, the numbers are still too high.

Image of Salvador Dali, who inspired the name of DALL E 2
Salvador Dali, the Spanish artist who painted the famous ‘Persistence Of Memory’


How was it created?

A CLIP (Contrastive Language—Image Pre-training) model was used to train it. It is a neural network trained on (image, text) pairs. Roughly, around four billion (image, text) pairs were used to train it. It has a “zero-shot learning” ability (which is when a model attempts to predict objects that weren’t part of the training data) which makes it excellent to use.

How does it generate the images?

It uses a text description to create realistic and original images from existing images. It can also create variations of existing images by changing the colours, texture, dimensions, etc. A method known as diffusion is used to generate them. Diffusion is basically a process that gradually builds the image from scratch, adding elements and altering the image until a perfect image is formed. To make it more specific, it uses a fixed Markov chain to add noise to the data. The image is transformed to Gaussian noise and then reversed back to the original image. This traversing causes new data to be formed. Diffusion consists of two processes: forward and reverse.

Stunning works of art in minutes!

DALL E 2 has impressed all of the world by its capabilities of generating accurate and realistic art. Be it a human/living thing or a rendition of a scenery, it’s scarily realistic. And versatile, too. It can create art with specific angles, lens flares, perspectives, art styles, filters, impressions and much more. It truly is an art-lover’s heaven. While it’s not exactly known how long it takes to create them, the DALL E 2 mini approximately takes about 3 minutes to produce the desired output.

Even now, the DALL E 2 is being actively tested and learned, as its creators have admitted that they themselves don’t know fully about its capabilities.

There are some setbacks, too. For instance, people feel that this will only diminish the natural human creativity and leave artists jobless. Another thing is that it takes a while for people to properly learn about its limits. As a user, it may not yield the exact results while using it unless one learns how to word the descriptions properly to use it effectively.

As someone who enjoys the mere existence of art and who cannot draw or sketch for the life of it, DALL E 2 makes me happy. Let’s hope that this technology creates a revolution or even a movement in the world of art while not affecting artists.

To play a wordsearch game based on the blog, click here!

Click here to read about Starbucks

Click here to read about the cardiovascular system