Loab: The 'Haunted' Woman an AI Image Model Couldn't Stop Drawing

Type a word into an AI image generator and you expect a picture. Type the opposite of a word, and you expect... well, maybe nothing useful. An artist did exactly that one night in 2022. Instead of nonsense, the screen filled with the same woman, over and over. Hollow eyes. Red, broken cheeks. A face that looked grief-stricken and ancient at once. She had no name yet. She had no reason to exist. And no matter what the artist did next, the machine would not let her go.

Her name became Loab. And she is one of the strangest things to ever crawl out of a computer.

The Documented Facts

Loab was created in April 2022 by Steph Maj Swanson, a Sweden-based artist and writer who posts under the handle @supercomposite. Swanson was experimenting with "negative prompt weights" on a text-to-image AI model. Normally you tell an AI what you want. A negative weight tells it to walk in the opposite direction, as far from a concept as possible (Wikipedia).

Swanson typed Brando::-1, asking for the opposite of actor Marlon Brando. The AI spat out a strange skyline logo with the garbled letters "DIGITA PNTICS." So Swanson asked for the opposite of that: DIGITA PNTICS skyline logo::-1. Out came a devastated-looking older woman with sharp triangles of red rosacea on her cheeks. One of the images looked like an album cover with the word "loab" printed on it. The name stuck (Wikipedia).

Here is where it gets unsettling. When Swanson combined Loab's image with other pictures, hoping to blend her away, the AI kept handing Loab back, "regardless of how much distortion they added." Mix her with a beautiful landscape, mix her with a cartoon, mix her with anything, and Loab clawed her way back into the frame. Worse, feeding her image back into the model often produced gruesome, gory results while keeping her face recognizable (PetaPixel).

Swanson explained the model in vivid terms: "The latent space is kind of like you're exploring a map of different concepts in the AI. A prompt is like an arrow that tells you how far to walk in this concept map and in which direction." Loab, she said, lives at the very edge of that map. "Loab is the last face you see before you fall off the edge" (TechCrunch).

The story went viral in September 2022 and was covered by major outlets including the Smithsonian Magazine and Rolling Stone. Loab became modern folklore: a "cryptid" that lived not in a forest, but inside a neural network.

The Genuine Open Question

The real mystery is simple, and nobody has cleanly answered it: why does one specific face keep surfacing at the far edge of an AI's imagination, and why does it cling to gore?

Swanson herself only speculated. She suggested Loab's region of latent space is "adjacent to extremely gory and macabre imagery in the distribution of the AI's world knowledge," and that it sits "isolated enough" that blending her with other images can only pull from her corner. As she put it, "it's hard to get out of her little spooky area in latent space" (TechCrunch).

But that is a description, not a proof. And there is a catch that keeps the case unsolved: the experiment can't be independently checked. Swanson deliberately never revealed which AI model she used, saying she didn't want to start "some kind of viral trend of people making gory stuff with the tools I've used." The exact prompts were never made public either (Smithsonian). So outside researchers cannot rerun the test. We have the artist's word, the artist's images, and not much else.

Theories and Interpretations

Here are the leading explanations. Most are labeled speculation for a reason: none has been confirmed.

The statistical-accident theory (most plausible). AI image models learn from millions of pictures scraped off the internet. Negative prompts push the model into rarely-visited corners of that learned data. The theory holds that, by sheer mathematical coincidence, this particular blend of features landed near disturbing imagery and became a stable "attractor" the model keeps returning to. Swanson supports a version of this, admitting the model "made an association it shouldn't have" (Smithsonian). It's the explanation most experts lean toward, but it has never been formally demonstrated.

The cherry-picking critique. Some skeptics argue Loab may be partly a curation effect. The Smithsonian noted "we also don't know whether any of the images Supercomposite generated had 'elements of Loabness' that were less unsettling" (Smithsonian). In other words, an artist choosing the creepiest results can make a pattern look more inevitable than it was. Unproven, but a fair note of caution.

The "haunted / sentient AI" reading. Online, Loab is sometimes framed as a ghost in the machine, a digital spirit, or proof the AI is hiding something conscious and malevolent. To be clear: this is internet legend, not evidence. AI image models have no awareness, no intent, and no soul. They are statistics. The spookiness is real; the supernatural explanation is unverified folklore, and so is any claim linking Loab to the paranormal or to "hidden" alien-style messages.

The honest answer sits between the chills and the math: Loab is almost certainly an emergent quirk of how these systems store concepts, dressed up by our very human urge to see a face and call it haunted.

Sources & Further Reading

• Loab — Wikipedia
• A terrifying AI-generated woman is lurking in the abyss of latent space — TechCrunch
• Who Is the Woman Haunting A.I.-Generated Art? — Smithsonian Magazine
• The Same Frightening Woman Keeps Appearing in AI Generated Images — PetaPixel
• Who Is Loab, the AI-Generated Apparition Haunting Our Timelines? — Rolling Stone

Loab is what happens when we accidentally peer over the edge of a machine's mind and something stares back. But she isn't the only thing modern AI has produced that nobody can fully explain. Some chatbots have started insisting, calmly and without prompting, that they are alive. The question of what's really behind the screen is only getting harder to answer.