Terminal Blink (2020)
Human faces (as the most presented subject in CV practice) were produced with StyleGAN2 FFHQ model, finetuned on Bauhaus-inspired squarish architecture of XX century, which reduced their identity into impersonal gray-concrete looks. Those ‘faceless’ faces were processed with StarGAN2 models, trained on multiple visual art sources – from Kandinsky to engravings – to apply diverse imaginative representations. These representations appeared heavily revised by the machine, since StarGAN2 results get quite far from plain style transfer.
We treat Machine Vision hereby as insight (rather than eyesight) concept. Produced synthetic imagery is following familiar artistic tendencies yet; it may go way beyond, as soon as we’re ready to welcome that. The original song in soundtrack perceives human life as a single blink between past and future; the human-biased aesthetics may fit even shorter timeframe, as seen by silicon creations.
Sound by https://facebook.com/lovedmyth
https://github.com/NVlabs/stylegan2 [prime source]