I liked it (unusually enough for some random blog) so I'll volunteer my take on it.
Remember how Stable Diffusion works? It's essentially noise reduction cranked up to eleven. You train a neural network to clean up slightly noisy images. When it gets good at it you turn up the noise a notch and repeat. You keep doing that until the network will look at pure noise and "repair" it into an image of whatever.
Animal brains have to do something similar, because our senses suck. Take our eyes; the part of your field of view which is actually in focus is tiny [1], so you are constantly updating an internal 3D model of the world, one small patch at a time [2], interpolating between patches and editing out things like your nose to replace it with a hallucination of whatever it happens to be occluding (except right now, because I reminded you of it).
In this sense, what you "see" is not so much the real world as a hallucination guided by your sensory input.
What happens when you reduce the information content of that sensory input, either by making it "unstructured" [3] or by removing it altogether [4]? Same thing that happens to Stable Diffusion when you present it with pure noise: it hallucinates freely.
So it makes sense that lying down on a comfortable bed in a dark, quiet room and shutting your eyes - all steps toward sensory deprivation - should have a similar effect.
