Image via Shutterstock
Enter another addition to the ever-growing list of things that could become fully automated, including
nails and
boba: art. Specifically, pieces that can be created with just a simple phrase or sentence. Even more specifically, pieces created using image-classifying software enhanced by hackers.
In January, OpenAI announced the introduction of an image-classifying AI, ‘CLIP’. This was the same company behind GPT-3, which was rolled out in May last year, which had the ability to generate extremely human-like text just from a simple prompt. CLIP was designed to match image to text; so entering the word “book” would deliver an image of a book.
OpenAI describes it as “part of a group of papers revisiting learning visual representations from natural language supervision in the past year.”
However, some saw greater potential in the software. This March, Ryan Murdock, a machine learning artist, engineer, and hacker, managed to successfully connect CLIP and another AI, VQ-GAN. He then worked on it until it could produce the high-quality images that it does today.
GANs, or generative adversarial networks, were the most powerful image generation tools available to the public previously. After the networks undergo “training,” they can generate new images that are similar to those they were trained with. However, they’re unable to generate images from prompt alone, which is where CLIP comes in.
Twitter user @images_ai shared a tutorial on how to use the combination of software. Ever since the art and machine learning communities caught wind of Murdock’s discovery, a large collective of slightly surreal, but still stunning, 3D work has been produced.
[via
VICE, cover image via
Shutterstock]