This art is made out of pixels

#draft

They are all pixels

Unless it is printed like the [[ Self Portrait, 1659 | Rembrandt ]], none of this work exists in the physical world, and the operation of a text-to-image engine remains indifferent to the existence or rules of the physical world. The “accuracy” of a model is in aligning text to pixels, not pixels to the outside world. The spectrogram example is telling – the AI doesn’t know or care at all that this corresponds to music when it diffuses one into being. There is no reason that it would know any more how the parts of a jet engine are supposed to relate to each other. What we get are photographic surfaces that are not tightly coupled to a logically consistent underlying structure or set of ideas.

This is the opposite of what a broad segment of paintings are meant to do. Paint is sloppy, simultaneously low and high-resolution, and a three-dimensional physical substance with complex and subtle properties. It is used by many painters as an opaque layer to be placed over a human-readable notion of physical space. Like a lumpy blanket, it obscures some things and accentuates others. Its inaccuracies contribute a productive randomness which is as much a part of the painter’s pallet as mineral spirits, and a large part of painting practice is learning how to wield the stochasticity of paint without extinguishing its power or your intentions. Painting is about getting things right and wrong simultaneously. You can be right, wrongly, and wrong, rightly.

obvs connection to smart dumb

connects to “getting things wrong” and Eufert Latrine – the jump between GANs and the current gen.