Although anyone can play a zombie on television or dress up as one for Halloween, literally becoming a member of the undead is still impossible. It is possible, however, to use machine learning to take your face and make a solid prediction about what it’d look like if it were zombified. And, if you can handle seeing your own mug dripping with flesh, you can do exactly that on the website, Make Me a Zombie.
Computer vision engineer Josh Brown Kramer recently posted to Twitter some of the results Make Me a Zombie can deliver. Kramer, along with others, created the site, which utilizes AI to transform users’ faces into their zombie one-offs.
In regards to the AI, Kramer et al. cultivated a machine-learning algorithm to make the horrifying facial transformations. Kramer wrote in a Reddit post that he utilized a generative adversarial network, or GAN in order to take users’ faces and, in effect, make predictions as to what they’d look like as zombie ones. “Zombie,” in this case, being defined for the algorithm via 300 images of zombie Halloween masks and people in zombie makeup. (There was obviously a lot more to the process of creating the GAN, which is explained in more detail in the post.)
Happy to report it works on babies.— Laura Kajpust 👩💻🎨 (@delusioninabox) October 25, 2020
Although...it definitely added a lot of teeth she doesn't have. 🤣 pic.twitter.com/kM7xdQIoar
The intricate technical details of the AI algorithm aside, Make Me a Zombie is, most critically, quite accurate. Unlike many other AIs that transform faces, Make Me a Zombie doesn’t lose the original face in transition. I.e. if you were to show your zombified face to a family member, they’d know it’s you in a heartbeat.
On top of being accurate, Make Me a Zombie also has pretty darn good taste. The style with which the AI applies “zombiness” to a face is a perfect blend of macabre and goofy. Plus, it makes some people look kind of hilarious, while it morphs others into absolute horrors. Apparently the AI even works on people already in zombie makeup, as you can tell from the picture immediately above.
Feature image: Steve Wilson