Spookycookie.co
Spookycookie.co is a project realised during a one-week machine learning workshop with Dario Cotti and Yllka Shehu, where we exploited artificial intelligence to generate random gingerbread images and Christmas-themed messages. We then created ten Christmas cards, each with a particular image and message, and different from the classic cards.
The workshop has been “an experiment of a creative collaboration between humans and machines. An experimental way of introducing algorithmic design in the creative process, while getting familiar with the underlying mechanics.”
After being inspired by some Victorian Era creepy Christmas cards, we decided to use machine learning to generate spooky visual imagery for the upcoming holiday season.
We chose to use gingerbread men as the main character for our cards because we wanted to have an easily recognisable element, but also because homemade cookies can also turn out imperfect and weird.
We then built a dataset of 418 images, all cropped into a square format, and trained a model using StyleGAN2 with 3000 steps in RunwayML. After the training was done, we selected 3 generative videos and around 100 images out of 800 that we generated.
We then built a dataset of 418 images, all cropped into a square format, and trained a model using StyleGAN2 with 3000 steps in RunwayML. After the training was done, we selected 3 generative videos and around 100 images out of 800 that we generated.
As a final step, we used an existing GPT-2 model in order to generate Christmas messages for our cards, which were all based on a simple initial sentence: “Merry Christmas and...”
