Metro

MIT’s ‘Nightmare Machine’ transforms photos into creepy images

You no longer need to step into a haunted house, or watch a cinematic thriller, to conjure up a good scare.

As a treat leading up to Halloween, researchers at the Massachusetts Institute of Technology’s Media Lab launched the “Nightmare Machine,” a project that uses deep-learning algorithms — or artificial intelligence — to generate “zombified” portraits of people, and transform pictures of recognizable landscapes into haunting imagery.

Manuel Cebrian, a principal research scientist at CSIRO’s Data61, in Australia, and one of the project’s creators, said the Nightmare Machine follows in the tradition of other MIT “hacks,” or pranks, that the school is known for.

Advertisement

Cebrian said the group wanted to playfully commemorate humanity’s fear of artifical intelligence, which he believes is a growing theme in popular culture — especially movies.

Get Fast Forward in your inbox:
Forget yesterday's news. Get what you need today in this early-morning email.
Thank you for signing up! Sign up for more newsletters here

“We found it appropriate to explore how machines, themselves, can generate the scary content,” he said in an e-mail to the Globe. “So we launched the Nightmare Machine.”

As part of the plan to creep out the public, the team put together a website showcasing the “haunted places and haunted faces” created by the deep-learning algorithms.

On the site users can watch well-known landmarks like the Taj Mahal, Golden Gate Bridge, and Statue of Liberty go from picturesque places to post-apocalyptic destinations after different styles are applied to the images.

Users can also vote on a series of creepy photographs showing nightmarish faces.

Advertisement

“The algorithm first learns how a face looks like — this is done by feeding the algorithm 100,000 celebrity faces — and then the algorithm is able to ‘generate’ new faces on its own,” explains Pinar Yanardag, a postdoctoral student at the Media Lab. “After we generate such unique new faces, we ‘zombified’ these faces by teaching another algorithm how a zombie face looks like.”

Voting on the scariest faces will help the algorithm better learn what really makes people’s hair stand on end.

The group is close to hitting 1 million visitors to the website in just under a week.

The group also has an Instagram page featuring pictures of weddings and Disney characters that have been mutated into dark and ghastly images. Each is pocked with black swirls and what look like tiny skulls.

Even though the project is tied to Halloween, Iyad Rahwan, an associate professor at the Media Lab, and one of the creators, said the main goal of the Nightmare Machine is to understand the barriers between human and machine cooperation.

Advertisement

“Psychological perceptions of what makes humans tick and what make machines tick are important barriers for such cooperation to emerge,” he said in an e-mail. “This project tries to shed some light on that front — of course in a goofy, hackerish Halloween-manner.”

Steve Annear can be reached at steve.annear@globe.com. Follow him on Twitter @steveannear.