It's fun to see the different ways people are exploring image synthesis with OpenAI's CLIP. This one uses a differentiable drawing library, with some great results. I've been having a ton of fun messing about with it - highly recommend testing it for yourself :)
My experiments:
https://twitter.com/JohnowhitakerA/status/141012787813467750...
My favorite is seeing what it does with more abstract ideas ('fear and loathing' or 'The face of God').