THE ARTISTS OF THE FUTURE
Artificial intelligence (AI) is not only changing economics and healthcare. The technology is also making its way into creative activities – the best example? AI-generated art. People around the world are currently creating AI art memes and taking social media by storm. An overview of the most capable art generators and the most memorable memes.
It is one of the trends, if not THE trend, taking social media by storm this year – art created by an Artificial Intelligence (AI). In June, the AI-generated Cosmopolitan cover made headlines, in July, the world's latest selfie sent shivers down its viewers' spines and now TikTok has introduced a filter based on a simple text-to-image generator. There's no doubt about it: artificial intelligence is taking over the world of creative processes.
While the most popular of these generators mainly satisfy the desire for fun and entertainment, Google, OpenAI & Co are tinkering with more elaborate tools: they generate quite photorealistic images through text input and could trigger a real creative revolution. These text-to-picture models are not (yet) freely accessible or free of charge, but offer a test environment plus detailed documentation that one can delve into:
Dall-E (the name is derived from Pixar's favourite robot Wall-E and surrealist painter Salvador Dali) is a 12-billion-parameter version of GPT-3. It is trained to generate images from text descriptions using a dataset of text-image pairs. The second version can create high-resolution, photo-realistic images.
On the website there is the possibility to create your own images from given text snippets.
The impressive technology is explained clearly in the following video:
Imagen is a text-to-image diffusion model with an impressive degree of photorealism. Imagen builds on the power of large transformer language models in understanding text and draws on the strength of diffusion models in highly realistic image generation.
The Imagen website currently offers the possibility of a "sneak peek" in the form of demonstration examples and detailed descriptions of the process.
If you like, you can also delve into the research paper that has been made available (46 pages, as of May 2022).
A successful video contribution by 2 Minute Papers shows how impressive Google's text-to-picture technology already is:
Maybe one day these text-image generators will spare companies the need to go to a creative agency, when a visualisation from a generator will be completely sufficient. Maybe not only will an AI artwork be worth a lot, but one day the art market as a whole will be completely changed. Perhaps it will no longer be possible to distinguish what is real and what is not, and the world will come apart at the seams. Or there will be a new, more regulated internet ... or maybe not. So far, only one thing is clear: the visuals generated by these AI programmes are making the internet a more entertaining place through numerous memes (what else?).
Below is a small list of well-known and strange AI creations, with links to the tool used in each case:
THE ASTRONAUT ON THE COSMOPOLITAN COVER
For the cover of their "A.I. Issue", the lifestyle magazine Cosmopolitan relied on Dall-E's skills. The text input was: "a strong female president astronaut warrior walking on the planet Mars, digital art synthwave" – about right, or what do you think?
THE LAST SELFIE ON EARTH
... was published by TikTok user Robot Overloads published – who also used Dall-E 2 for this, with the titular text input. The result is rather disturbing in its various versions. In any case, not for optimists.
ANIMALS THE LIKE WE HAVE NEVER SEEN BEFORE
Google's Imagen surprises with almost unseen photo realism. No matter how unrealistic the text specifications, the generator does not seem to reach its limits. But see for yourself:
Text input: "A giant cobra snake on a farm. The snake is made out of corn."
Text input: "A photo of a Corgi dog riding a bike in Times Square. It is wearing sunglasses and a beach hat."
OLD AND NEW MASTERS
Even real works of art, especially Dutch ones, are not safe from the robot artists ...
Text input: "A sea otter in the style of Girl with a Pearl Earring" (created with Dall-E / OpenAI)
However, there are also numerous tools that are already available to users on the internet for free or for a fee - and are also used extensively. The Night Cafe platform gives the user various options for proceeding: In addition to the text-to-picture model, there is also the "style transfer", which applies the style of one picture to the other. In the above picture, a famous photograph of the moon landing was transformed in the style of one of the most famous paintings by the Dutch master Van Gogh.
The Deep Dream Generator uses a similar system and merges existing images, also with impressive results, generated by numerous users around the world.
But, back to the artists one more time...
This is probably what would have happened if the impressionist painter Claude Monet had found water lilies and sunrises boring and had instead been more inclined towards palaeontology. These pictures were created with the AI art tool Dall-E mini, which has become the quintessential meme generator.
The image generator is based on OpenAI's GPT-3 software and Google's TPU Research Cloud, so it accesses information collected in Google's information hubs.
It was already created in 2021 and has enjoyed great popularity since mid-2022 at the latest.
The Twitter channel Weird Dall-E Mini Generations collects the strangest creations:
You can hardly get enough, can you? Dall-E mini also allows a previously unknown insight into the Binary Dreams editorial team ...
DALL-E MINI x BINARY DREAMS
Whoever finds out which Trivadians are depicted in these pictures will get a free ice cream!