top of page

About

A.I. Generated NFT Artwork

I create Art Collections with various themes of images and Artificial Intelligence. By using GANs and collections of images to train models, the computer generates images that it considers similar to the originals. I then curate the generated images and NFT them to harness the power of blockchain, which provides transferability and authenticity of the artwork. 

What Are GANs?

Generative Adversarial Networks, or GANs for short, are an approach to generative modeling using deep learning methods. Generative modeling is an unsupervised learning task in machine learning where the computer learns and discovers patterns and regularities in the input data, a set of images, which "trains" a model. This model can then draw images that statistically could be from the original input data set. On the right, you can see the first NFT I created using a StyleGan2 model I trained using 1232 Abstract images.

Abstract image created using Artificial Intelligence and Machine Learning.
Images of people created using StyleGan that have never existed.

GANs train a generative model by using two sub-models: the generator model that is trained to generate new images and the discriminator model that trains to classify the examples as either fake (generated) or real (from the input image set). The two models are trained simultaneously until the discriminator is fooled about half the time, meaning the generator model is producing good examples. Training a model to create quality images takes a lot of computing power, time, and thousands of images to train with. For example, the images on the left are of people that, well, don't exist. The model to generate the images was done by Nvidia using StyleGan2, hundreds of thousands of photos of real people, and 8x Tesla V100 GPUs. It took about 9 days to train the model, and you can see more examples of people that don't exist.

The AI Artwork

The possibilities with GANs are endless. They can be used for facial recognition, image generation, creating Deep Fakes, Snapchat filters, and much more. What interests me the most as an artist is to see what the computer will come up with after days of training on various themed datasets. The results sometimes can be surprising. After training the dataset, I generate 1000 images, from which I pick 10 to NFT and mint into a collection (if you do not know what an NFT is, you can read more on them here). If I decide to further train the GAN or adjust the dataset in any way, I will release the results in additional Generations. These NFTs will soon be printed onto canvas using a robotics arm powered by AI, when that is completed, you will be able to claim your canvas.

On the right you can see the progression of a GANs training. The top row is what is produced as "fake" images when you first start the training, while the results on the bottom row are after a few days of training.

Shows the progression of how a machine learning algorithm, StyleGan, learns and produces images.
bottom of page