Start to think of the world as a sequence of choosing things based on probabilities. That's basically modern modern AI text, sound, and image generation. You somehow calculate logits, then sample them. Bonus points for interpreting cloud patterns, meaning to infer a similar known shape from noise, i.e. the denoising decoder like Google's imagen.

In effect, geberative art is about gently nudging noise distributions and about finding the valuable needle in a haystack of noise.