A number of experiments in audio-reactive, automated – and controlled – traversal of the latent space of a large generative adversarial network (in this case, the model known as BigGAN by Google Deepmind). These videos use many different methods of navigating the latent space. Each video uses a different form of audio reactive traversal, and some additionally use a custom controlled method as described here (e.g. in “I wanna be a dog”). Many BigGAN (or other GAN) music videos followed, but these are the very first – at least to be posted publicly. (The tweet thread can be found here).
#BigGAN Music Video Studies (2018)
#BigGAN take your brain to another dimension
Pay close attention.
Music: The Prodigy – “Out of space”.
It’s more fun to compute.
When I was a kid, I was a bit of an anti social loner, tucked away in my room. I didn’t go out much. People were like “why don’t you go out and play with the other kids?” And I was like “but it’s more fun to compute”. (I program my home computer, beam myself into the future).
Music: “Kraftwerk – Home Computer”.
I wanna be a dog
(or a door?)
Music: The Stone Roses – “I wanna be adored”.
#BigGAN Madness
Music: Iron Maiden – “Can I play with madness”.
Come to #BigGAN
Music: Aphex Twin – “Come to daddy”.
Related work
- Deep Meditations (2018)
- Deep Meditations: Morphosis (2019)
- Deep Meditations: Abiogenesis (2020)
- Deep Meditations: We are all connected studies (2020)
- Deeper Meditations (2021)
- BigGAN studies (2018)
Acknowledgements
Created during my PhD, funded by the EPSRC.
Model is pre-trained BigGAN by Andrew Brock, Jeff Donahue, Karen Simonyan from Google Deepmind.
Amongst many people, I’d like to especially thank Nina Miolane and Sylvain Calinon for their contributions and help with Riemannian Geometry – and particularly Miolane et al for geomstats.