Generating Art with VQGAN+CLIP AI

I’ve been playing around with OpenGAN, TensorFlow, torch and a few other goodies lately. Adding CLIP to the mix gives me the ability to use common phrases to guide the AI to creating works. It’s fascinating to see what is generated based on the training models used. Some are surprisingly good and others, well, I’ll let you be the judge. Here are a few of the pieces along with the phrases that generated them:

This next one is interesting. The first frame that was generated is this one:



However, the last frame rendered when I killed the generation was this one:


I believe that the training model includes some artifacts that pushed it this way.

For this next one, I really liked the first rendering so I saved it separately.

And the fifth frame was good, so I stopped the AI there.

All of these are NFTs for sale over at Opensea, so https://opensea.io/collection/gan-gen-xeries-3 if you’re interested in buying one.