After discussing the AI-generated art with a friend from Italy I decided to generate some more images in the style of Leonid Afremov. His style takes my breath away!
The generation of one image takes between 30 – 32 seconds on my MacBook Pro M1 Max. The generated images have a resolution of 512×512 pixels.
Here’s a sample image and I’ve uploaded more to a flickr gallery.
Past Intel Macs were very slow at most neural computation tasks and so I was excited to give the latest M1 Apple Silicon chip a try. According to pytorch.org the speedup is 8 – 21 fold over CPU training and evaluation.
Usually you take a computer with Nvidia GPU if you want to train your own AI models at home. For a nomad like me that meant a gaming laptop with a GeForce 1070 graphics processor. The disadvantage is clear: it’s 4 – 5 kg of unnecessary hardware to carry around. So I had a heavy incentive to find a different solution that would allow me to train AI models without carrying a second laptop even in cases where the internet connection is slow or non-existent.