Image-generating artificial intelligences have become popular in recent months. Since Dall-E and MidJourney were opened to the general public, the Internet has been filled with all kinds of artistic creations. Thanks to its simplicity (you just have to write a prompt), these machine learning models trained with billions of images, are capable of creating amazing works in a few seconds.
And of course, they have not been long in coming the first urban legends based on AI concepts. Basically because few can explain what leads these intelligences to create what they have created. In the end, it is complex neural networks that carry out these images and there is always part of the mystery. That is, they are unpredictable results. An example is the case that has been tormenting the artist Supercomposite. Some days ago, published on Twitter a series of disturbing images of a woman who appears by asking the AI to create certain words.
She is “Loab”.
It was actually first discovered as a result of a technique called “negative prompt weights,” in which a user tries to get the AI system to generate the opposite of what you write in prompt. That is, it is capable of generating what it thinks is the opposite of the message you indicate. In this case, by giving her the word “Brando”, she generated the image of a logo with a city skyline and the words “DIGITA PNTICS”. When the artist used the same technique Hoping that the opposite of those words was an image of Marlon Brando, Loab appeared.
All images were from the same evil-looking older woman with defined triangles of rosacea on her cheeks. “Since Loab was discovered using negative prompt weights, it is made up of a collection of traits that are equally far from anything. But their combined traits remain a cohesive concept for the AI, and almost every related image contains a recognizable Loab “explained the artist.
The images went viral on networks, which generated all kinds of speculation about what could be causing this. The scariest thing is that the artist’s instructions to the AI continued to generate images derived from Loab’s original that deviate almost always towards horror, graphic violence and gore. No matter how many variations have been made, they all seem to feature the same emaciated woman.
“Angel’s Hall + Loab too produced images with so much blood that probably very few people want to see them. I don’t feel comfortable posting the most disturbing, images almost snuff of dismembered and screaming children. Through some kind of emergent statistical accident, something about this woman is adjacent to extremely gory and macabre images in the distribution of the world’s knowledge of AI,” Supercomposite noted.
The AI can hold on to Loab’s idea so well that it can persist through generations of this kind of crossbreeding, without using the original image. Here’s Loab as Kirby, as a bee or celebrating Pride. praise can be recognizably translated into many genres and contexts. “Even when her red cheeks or other features disappear, the ‘loudness’ of the images she’s in is undeniable,” he noted.
Images: @supercompositecreated with AI.