Meet 'Loab', A Nightmarish Old Woman Who Seems To Haunt An AI-Image Generator

Artificial Intelligence can be trained to do anything. But the results the AI create depend on the training materials, the algorithms, and more.

But depending on the various criteria and factors, some AIs do better in certain thing than others. In this case, one particular AI image generator can create eerie looking old woman that can only be found on horror movie or nightmare. And it seems that the woman is here to stay.

Her name is Loab (pronounced like 'lobe'), and she was "discovered" by a Swedish-based artist who goes by the name of Supercomposite on Twitter.

The user said that she can be 'summoned' using the right prompt inputted into an AI image generator.

The artist described the appearances of the woman as an unexpectedly emergent property of the software, saying that she was discovered when they asked the unspecified AI image generator to produce something "as different from the prompt as possible".

The internet who realized the existence of this woman, have started created artworks of her.

Now so well-known, the mysterious woman even has her own Wikipedia page.

Supercomposite is among the creators who explores the realms of text-to-image AI generators, in order to see what the emerging technology is capable of.

And in this case, Supercomposite said that the woman was discovered when the user was experimenting with negative prompts.

When Supercomposite asked an AI image generator to find the extreme opposite of something, the artist stumbled across an old woman who looks like a rag doll at first.

And when Supercomposite ran the prompt again, the same old woman come back.

But this time next, the woman's photo has the word 'loab' next to it.

"The AI reproduced her more easily than most celebrities. Her presence is persistent, and she haunts every image she touches," Supercomposite wrote on Twitter.

"Take a seat. This is a true horror story and veers sharply macabre."

Sometimes, results can get creepier and creepier, and even gory.

Among the many of Loab iterations Supercomposite has summoned, some showed how the old woman feature dismembered body, and more.

Some AI-generated images were so grotesque that the artist decided to flag them on Twitter.

"I was ripping Loab apart and putting her back together. She is an emergent island in the latent space that we don't know how to locate with text queries," the artist said on Twitter.

Supercomposite calls Loab as a result of an "emergent statistical accident."

Supercomposite speculated that something about the Loab's imagery that could be "adjacent to extremely gory and macabre imagery in the distribution of the AI's world knowledge."

But what intrigues the artist is that, the woman seems to stay.

Supercomposite said that when they combined images of Loab with other pictures, the subsequent results consistently return an image including Loab.

Regardless on how much Supercomposite used other images to distort the results, Loab's nightmarish figure continues to haunt the results.

Supercomposite speculated that the latent space region of the AI map that Loab is located in, in addition to being near gruesome imagery, must be isolated enough that any combinations with other images could only use Loab from her area and no related images due to its isolation.

It was only after enough crossbreeding of images and dilution attempts, Supercomposite was able to eventually generate images without Loab in them.

However, crossbreeding those diluted images would eventually lead to a version of Loab to reappear in the subsequent images.

A text-to-image model is a machine learning model which uses text as an input, in order to generate image(s) that match that description.

While such models were already developed in the mid-2010s, as a result of advances in deep neural networks, it was only in 2022, that state of the art text-to-image models, such as OpenAI's DALL·E 2, Google's Imagen and Stable Diffusion began to the public realize of their existence.

This happened because the technologies are able to create quality images of real photographs and human-drawn art.

AI-powered image generators are trained using datasets of images and photos, all based on human knowledge, culture, and traditions of art.

What this means, humans should be able to replicate whatever AIs can do.

The thing is, researchers don't really know how AI's 'brain' work.

The so-called black box has made the AI field both fascinating and complex.

And Loab here, is not just a creation of technology, but also a reflection of what the technology can do.

Some suggest that Loab is a creation of a woman with violent imagery, and that social bias in the data was used while modeling the AI.

Because many have referred Loab as the "first AI-generated cryptid," the woman has gone viral on social media.

While there a number of discussions have been made to find whether Loab is just a legitimate quirk of AI art software, or a cleverly disguised creepypasta, in reality, people seem to like the creations.

Supercomposite admitted that "Loab isn't really haunted, of course", but since more images of Loab are being made by other artists now, Loab has become part of the internet landscape.

The boundaries of artistic practice are expanding, and Loab is just a start.

"If we want to get rid of her, it’s already too late," Supercomposite said.

Supercomposite said that "for various reasons" they are declining to disclose the software used to create the images.

They did "not want to advertise" the AI tool, and didn't want to "start some kind of viral trend of people making gory stuff with the tools I've used."