With the recent rise in Artificial Intelligence’s popularity, also I developed a great interest in the technical possibilities that came with it. This curiosity sparked a personal project where I decided to build my own AI during my minor. My goal was to get a better understanding of how AI tools can be used for creative and social purposes.
The core of this project involved building and training an AI with a huge collection of different flower pictures. I let it undergo several days of training, a process that continued mostly while I was asleep. I ended up with very abstract and vibrant coloured pictures that looked somewhat flower-like.
The magic behind this project lies in a framework called Generative Adversarial Network (GAN). It’s like a creative engine that can generate brand-new images. However, due to the limitations that came with the intensive process of training the AI, the images the AI generates are a modest 56x56 or even 28x28 pixels in size. That means they lack a lot off finer details, but it also opens up a world of possibilities for future improvements and playful experimentation with the AI.
Consider this project a work in progress. The AI is able to learn from various types of images, making it a versatile tool for future creative projects.
Dataset images
