Blood-curdling AI-generated videos go viral on Reddit; this is something you can’t unsee

Stable Diffusion 3 Medium, Stability AI’s latest text-to-image model, is something out of a horror movie. Redditors have taken to the platform to mock the model and showcase their crazy creations.

We’re all aware that text-to-image artificial intelligence (AI) models aren’t very accurate when it comes to creating images of people. But this new model is something else.

Stability AI, known for its text-to-image model Stable Diffusion, has recently released its latest model, Stable Diffusion 3 Medium, as part of the Stable Diffusion 3 series (SD3).

Despite being described as Stability AI’s “most advanced text-to-image model yet,” the results are wildly inaccurate and deeply harrowing.

Users have taken to Reddit to showcase some of the deeply disturbing images that SD3 Medium has created.

One user posted a collection of SD3 Medium body horror with the caption simply stating, “I’m sorry…”

This blood-curdling collection of videos shows a woman with four fingers and incredibly tiny arms opening her disjointed mouth to let out what can only be assumed as a harrowing scream.

The video then cuts to a woman whose hands constantly oscillate between having four and five fingers on each hand. What lies behind her are dismembered, unmoving arms discarded in atomic green grass. Oh, and the woman doesn’t have any legs.

Arguably, the most disturbing image generated by SD3 Medium is of a blonde “woman” lying in the grass. Doesn’t sound too horrifying.

However, she has three short legs that appear to have hands for feet accompanied by an odd number of toes. Her squashed face is upside down and missing features.

I tried out a Demo of Stable Diffusion 3 Medium, which is available on Hugging Face. I prompted the AI model by saying, "I want an image of a woman laying on the beach," and this is what I got. Horrifying.

In the thread, “Is this release supposed to be a joke? [SD3-2B],” many people have complained about Stability AI’s “most advanced model” and its inability to create anatomically correct images.

According to Stability AI, the new model “comprehends long and complex prompts involving spatial reasoning, compositional elements, actions, and styles.” Yet, it cannot reasonably generate an image of a person with an appropriate amount of fingers.

One Reddit user said, “I haven't been able to generate a single decent image at all outside of the example prompts. I've tried highly descriptive prompts with no luck. Even an absolutely basic one like "photograph of a person napping in a living room" leads to Cronenberg-esque (a reference to Canadian film director Cronenberg who originated the body horror genre) monstrosities.”

The user's image shows a “person” lounging on a couch with their “leg” on a side table. However, it's impossible to tell where the leg starts and ends, as the “person” in the picture seems to have four legs and one foot.

Many users have speculated that Stability AI has filtered out all things deemed not safe for work (NSFW), which is why the model is creating nightmare fuel.

One user said, “It works fine as long as there are no humans in the picture. I think their improved NSFW filter for filtering training data decided anything humanoid is NSFW.”

Another user supports this statement by saying, “Believe it or not, heavily censoring a model also removes human anatomy.”

Stability AI states that it believes in responsible AI practices and has tested the model internally and externally to prevent harm. Yet, generating this kind of body horror has left many users scarred.