The same creepy woman keeps popping up in AI-generated footage
A terrifying demon called Loab continues to appear in images created by artificially intelligent (AI) text-to-image generators.
Loab was created by negatively weighted prompts, and her gruesome appearance constantly shows up in footage, even when the AI is directed away from Loab’s prompts.
The scary monster was discovered by Supercomposite who is a Swedish musician.
🧵: I discovered this woman, whom I call Loab, in April. The AI replicated her more easily than most celebrities. Her presence is persistent and haunts every image she touches. CW: Sit down. It’s a true horror story, and turns heavily macabre. pic.twitter.com/gmUlf6mZtk
— Supercomposite (@supercomposite) September 6, 2022
“I discovered this woman, whom I call Loab, in April. The AI replicated her more easily than most celebrities. Her presence is lingering and it haunts every image she touches,” writes Supercomposite.
What are negative weighted prompts?
Negative weighted prompts are when you tell an AI to make an image look as different from the text prompt as possible.
“With those [negative weights]instead of creating an image of the text prompt, the AI tries to make the image as different from the prompt as possible,” Supercompste explains on Twitter.
Initially, Supoercomposite tried to create a negative image of Marlon Brando, but instead an image of a cake was created.
“I wondered: is the opposite of this logo, in turn, going to be a picture of Marlon Brando?” Writes the AI image creator.
She then typed in the text that appeared on the cake with the negative weight instruction and received a set of “off-putting images”.
“All from the same devastated-looking older woman with defined rosette triangles on her cheeks.”
Supercompsite then used Loab’s photos as an image prompt, and to its horror, the AI continued to throw up bloody and extremely disturbing images.
She “cross-referenced” different images with Loab’s photos and time and time again, a distinctive and unsettling character from Load always appeared.
“There is something moving for me in these grotesque scenes and the despair, panic and sadness they convey,” adds Supercomposite.
“The images that result from the crossing of Loab with other images can in turn be crossed with other images. The AI can latch onto Loab’s idea so well that it can persist through generations of this kind of crossover, without using the original image.
As an example, Supercomposite posted a photo of Loab as Kirby, as a bee, and as an Avatar character.
“The concept of ‘Loabness’ has become more abstract for me. I would include it in prompts that I knew would distort it almost beyond recognition,” says the Swede.
“After disappearing from the picture breeding line, she sometimes reappeared, later in the line, out of nowhere.”
The creator of the image says the lesson from Loab is that the image prompt, rather than the text prompt, can be used as a custom vector “to interrogate latent space.
As if Loab wasn’t terrifying enough, this week also saw a viral video of Scott Lighthiser who used Stable Diffusion, a video transformer and an audio transformer to create the chilling video below.
— Scott Lighthiser (@LighthiserScott) September 7, 2022
The video, made exclusively from AI technology and using all-free platforms, highlights how fast this technology is growing as people start to exploit it in more advanced ways.
Picture credits: All images by Supercomposite.