AI Halloween avatars! StyleGAN2 Generator reveals your inner zombie

Would you like to see what you would look like as a zombie? Forget the makeup; Now there is a GAN for it. The popular StyleGAN (Style Generative Adversarial Network) is a GAN architecture extension made available by Nvidia in 2019 as an open source version that can produce stunning photorealistic images while giving the user control over the image style. This year’s new and improved StyleGAN2 has redefined the state of the art in imaging – and also inspired a range of fun and creative activities with faces.

StyleGAN technology inspired last month’s Toonify Yourself viral website, created by a few independent developers, that turns selfies into adorable big-eyed cartoon characters. Just in time for costume season, another indie developer brought facial image transfer technology to the other end of the cuteness spectrum and built a zombie generator.

The developer of “Make Me A Zombie” is Josh Brown Kramer from Nebraska, who has set up a website where anyone can upload pictures and use the generator for free.

Kramer explains that he broadcast a StyleGAN2 zombie generator first, then inspired by the Reddit post Cross-model interpolations between 5 StyleGanV2 ​​models – Furry, FFHQ, Anime, Ponies and a Fox model, created a hybrid StyleGAN2 model. The first layers of the model are from the original human image generator, while the last layers are from the zombie generator. Finally from the paper of the Yandex and the Moscow Institute of Physics and TechnologyStyleGAN2 distillation for feed-forward image manipulationHe used 50,000 image pairs (from the human StyleGAN2 generator and the zombie generator, respectively) and Pix2PixHD to learn how to map between image pairs.

There are two slight differences between the Toonify Yourself and Make Me A Zombie approaches:

• The zombie generator uses crappify to enhance the input data and intentionally introduces artifacts for resizing and compression.
• While the Toonify Yourself generator focuses on the texture of its comic book images, the proposed hybrid zombie model instead emphasizes the shape and orientation of the original image.

The system was trained on a hand-filtered zombie dataset, mainly collected by Pinterest and Google, that includes around 300 images of people with zombie makeup and zombie Halloween masks. Kramer says that by adjusting the stack size and learning rate, he could train the model on his Nvidia 2080 Ti GPU at home at a size of 1024 × 1024 in about a day.

The seasonal offer is full of social media, with scary and amusing results that even come from adorable babies.

To try it out for yourself, visit the Make me a zombie Website.

---

Analyst: Hecate He | Editor: Michael Sarazen; Yuan Yuan

---

Synchronized report | An Overview of China’s Artificial Intelligence Solutions in Response to the COVID-19 Pandemic – 87 case studies from over 700 AI providers

This report offers a look at how China has used artificial intelligence technologies in the fight against COVID-19. It’s also available on Amazon Kindle. Along with this report we also have one Database Another 1428 solutions for artificial intelligence from 12 pandemic scenarios.

Click here for more reports from us.

---

We know you don’t want to miss any news or research breakthroughs. Subscribe to our popular newsletter Global AI Weekly synced to get weekly AI updates.