Step 1: Base Layers
Traditional artist, @wheatatreat, creates a large set of individual faces and backgrounds.
Step 2: Dataset
Large dataset is created by scrambling up all of the faces and backgrounds and adding variations.
Step 3: Training
Train AI models (GANs) on the dataset created in step 2.
Step 4: (lots of) Image Processing
The models output lots of images that were visually similar to one another. This step compares all and removes similar images.
Step 5: Traits
After hand labelling some of the outputs, a convolutional neural network was trained to learn how to assign traits to the rest of the images.
We also used a pre-trained CLIP model in order to assign some additional traits.