Abstract Faces
Sixth NFT Collection
This collection is the sixth overall separate theme, Abstract Faces. I wanted to see what kind of results Artificial Intelligence would produce after learning on various images of abstract faces that are different in art style, expressions and colors schemes. The First Generation was created using StyleGan2 ADA machine learning algorithm that trained on 1980 images for 9 days. The Second Generation was trained on 4207 images that were mirrored on the Y-axis, increasing the dataset to 8414, for . Future generations will be created using various techniques, algorithms, and hardware that become available, while increasing the number of images used for training.
Each NFT comes with a code that is unlocked after purchase that you will be able to redeem and have a canvas print of the NFT shipped to you. I am currently traveling but when I settle down I plan on opening a physical NFT art gallery. Due to this, expect delays of receiving the printed NFT Canvas of 1 - 2 months, depending on your location. The price of the NFT incorporates all shipping charges. Each NFT is unique with only one ever created (minted).
Gen 1
Image Dataset #: 1980
Image Size: 1024 x 1024
Seeds : 1 - 1000
Total NFTs: 30
Minted: 30 / 30
Training Time: 9 Days
The first generation was created by training StyleGan2 ADA on 1980 images of abstract faces with many distinct art styles, expressions, and color schemes. The results came out amazing for a relatively small dataset, with many different varieties of images. It was hard to narrow down to just 30 to NFT, which is already more then I typically pick. There was also no issues such as mode collapse or overfitting.
Gen 2
Image Dataset #: 4207 x 2
Image Size: 1024 x 1024
Seeds : 1 - 2000
Total NFTs: 10
Minted: 36 / 36
Training Time: 10 Days
Blockchain: MATIC
This is the second generation of the Abstract Faces Collection. I added additional images to the dataset and re-trained the StyleGan2 algorithm at a higher batch size, increasing it from 4 to 16. The increase in batch size and data set size seemed to create additional diversity in the results. Overall, I am happy with the results of the second generation!
Gen 3
Coming In The Future