Ridiculed Stable Diffusion 3 excels at AI-generated body horror

Zoom in / Image generated by AI using Stable Diffusion 3 of a girl lying on the grass.

On Wednesday, Stability AI released weights for Stable spread 3 medium, an AI-powered image montage model that turns text prompts into AI-generated images. However, its arrival has been ridiculed online, because it generates images of humans in a way that seems like a step backwards from other modern photomontage models like Midjourney or DALL-E 3. As a result, it can produce wild, anatomically incorrect visual atrocities. Easily.

A topic on Reddit titled “Is this version supposed to be a joke? [SD3-2B],“Details the SD3 Medium’s astonishing failures at rendering humans, especially human limbs like hands and feet. Another thread titled, “Why is the SD3 so bad at generating girls lying on the grass?“It shows similar problems, but for entire human bodies.

Hands have traditionally posed a challenge to AI image generators due to the lack of good examples in early training datasets, but recently, several image synthesis models seem to have overcome the problem. In that sense, the SD3 seems like a big step backwards for the photomontage enthusiasts who gather on Reddit — especially compared to recent Stability releases like the SD XL Turbo in November.

“It wasn’t that long ago that StableDiffusion was competing with Midjourney, and now it seems like a joke in comparison. At least our data sets are safe and ethical!” books One Reddit user.

AI image enthusiasts have so far blamed Stable’s Diffusion 3 dissection failure on Stable’s insistence on filtering out adult content (often called “NSFW” content) from the SD3 training data that teaches the model how to generate images. “Believe it or not, heavy censorship of models also leads to the elimination of human anatomy, so… that’s what happened.” books One Reddit user in the thread.

See also  Final Fantasy 16 is in the "final stages of development"

release Stable spread 2.0 2022 suffered from similar problems with accurately portraying humans, and AI researchers soon discovered that adult content containing nudity was also censored. Severely hampered The ability of an AI model to create accurate human anatomy. At the time, Stability AI had reversed course with SD 2.1 and SD XL, restoring some of the capabilities lost with the exclusion of NSFW content.

“It works fine as long as there are no humans in the image, and I think their enhanced nsfw filter for filtering the training data decided that anything human is nsfw.” books Reddit post.

Basically, any time a prompt touches on a concept that isn’t well represented in its training dataset, the image synthesis model will parse down its best interpretation of what the user is asking. And sometimes that can be downright terrifying.

Using a Free online demo From SD3 on Hugging Face, we ran the prompts and saw results similar to those reported by others. For example, the prompt “Man showing his hands” returned an image of a man holding two oversized hands back, even though each hand had at least five fingers.

Stability first announced the Stable Diffusion 3 in February, and the company plans to make it available in a variety of different model sizes. Today’s release is for the “medium” version, which is a model with 2 billion parameters. In addition to the presence of weights Available on face huggingIt is also available for trial through the company stability platform. Weights are available to download and use for free under Non-commercial license Just.

See also  New N64 Emulator plugin adds ray tracing, widescreen, 60 frames per second (and more) to classics like Zelda & Paper Mario

Shortly after its announcement in February, a delay in the release of the SD3 model weights led to rumors that the release had been delayed due to technical issues or mismanagement. Artificial intelligence stability as the company has fallen into disarray recently with resignation To its founder and CEO Imad Mushtaq in March and then a series of Layoffs. Immediately before that, three principal engineers – Robin Rumbach, Andreas Plattmann and Dominique Lorenz –Leave the company. Its problems go back even further, with news emerging of the company’s poor financial situation The relationship Since 2023.

For some Stable Diffusion fans, the failures at Stable Diffusion 3 Medium are a visible manifestation of the company’s mismanagement — and a clear sign that things are falling apart. Although the company has not filed for bankruptcy, some users have He made dark jokes On the possibility after seeing SD3 Medium:

“I think now they can go bankrupt in a safe and ethical way [sic] The road, after all.”

Leave a Reply

Your email address will not be published. Required fields are marked *