The beauty of failing in AI-driven typography.

Evolution in comparisons and approaches (including an almost fail-free solution).

Generative AI is getting better and better. CLIP had problems with realistic faces. Diffusion models struggled with hands. And most of the models couldn’t cope with typography.

All these shortcomings have their charm. Imperfection is the beauty of AI-driven art.

It’s all narrow AI: StyleGAN’s models could only imitate written language, not understand it. For GAN, these were just patterns dominating the strange culture of these humans.

Created with StyleGAN-Model trained on CD covers (via Artbreeder)

Such non-existent languages were fascinating for humans, as if we were entering another reality through a portal — into new realities, as in Tommaso Landolfi's “Dialogo dei massimi sistemi”.

I followed this fascination in my short film “Nyeshkerh”: