The Javen Number
Every image that can exist already does. What's left is finding out what that means.
Start with a single pixel. In the ordinary RGB encoding every screen uses, it takes one of 256 × 256 × 256 = 16,777,216 values. Now take a grid of them — 1080 by 1080, a modest square image, 1,166,400 pixels — and ask a simple question: how many different pictures can this grid show?
The answer is 16,777,216 multiplied by itself 1,166,400 times. Written out, it has 8,426,914 digits. It is not infinite. It is a specific, finite integer:
1.93472 × 108,426,913
I have carried this number around for more than a decade, since a night I couldn't stop thinking about a grid of pixels, and for lack of anything else to call it I call it the Javen Number. What kept me up was not its size. It was what membership in that set implies.
Because the set is complete. It contains every photograph ever taken at that resolution. Every frame of every film. The face of every person who has ever lived, at every age, from every angle, in lighting that never happened. The crime that was never photographed, photographed. The page of text that refutes this essay. Every image lost to fire and every image not yet made. All of it is simply there, each picture sitting at its own fixed position, the way the number 7 sits between 6 and 8.
No computer can help you
The first instinct is to render it — walk the set from frame zero (pure black) to the last frame (pure white) and simply look at everything on the way. The arithmetic is merciless. Suppose every atom in the observable universe — about 1080 of them — were a computer, and each rendered a trillion frames per second, and all of them had been running since the Big Bang. Together they would have covered roughly 10-8,426,800 of the set by now: a decimal point, millions of zeros, then a one. The bar has not visibly moved, and it never will.
The obstacle is not that the set is undefined or infinite — it is finite and fully specified. The obstacle is scale, and scale of this kind is not an engineering problem. It is a wall the universe itself cannot climb.
The address is the image
Here is the pivot the whole idea turns on. Each frame's position in the enumeration — its index, its address — can be written in base 256, one digit per byte. Lay those bytes across the grid and they are the pixels. The address and the image are not two things linked by a lookup table. They are the same bytes read two ways.
The set needs no storage, because rendering a frame and knowing its address are the same act.
This is also where a seductive fallacy dies. If every image already exists, couldn't we transmit pictures cheaply by sending only their positions? No — because the position is the picture. Specifying one frame among 108,426,913 takes exactly as many bits as the raw image. The pigeonhole principle allows no shortcut: you cannot give all the frames names shorter than the frames. People have chased this ghost before — a joke filesystem that "stores" files as offsets into the digits of π, a 1990s inventor who claimed to fit films in kilobytes — and the counting argument catches all of them. The map is the territory.
The enumeration idea itself has an honorable ancestor: Borges' Library of Babel, and Jonathan Basile's online realizations of it for text and for small images. Pixel Theory arrives at the same territory independently and builds different instruments on it.
Meaning is compressibility
If no frame's full name can be short, why does a photograph shrink to a fraction of its size in a file, while noise will not shrink at all? Because compression is not naming — it is pattern. A meaningful image is mostly redundant: skies are smooth, edges continue, faces are symmetric. A compressor exploits that redundancy. Noise has none to exploit.
This gives us something remarkable: a working, physical definition of meaning. Export a photograph's address from the explorer and it compresses seventy-fold. Export a noise frame's address and it compresses by a factor of exactly 1.000. The difference between meaning and noise is measurable, and it is pattern. Kolmogorov complexity, as a download button.
And the islands can be measured, not just felt. A counting argument puts a hard ceiling on meaning's share of the space: at most 256K frames can have a description of K bytes or fewer, simply because there aren't more short descriptions than that. A 1080×1080 photograph that compresses to 200 kilobytes therefore lives on an island of at most 10481,000 comparably-compressible frames — an unimaginably vast island that is nonetheless a 10−7,945,000 sliver of the whole ocean. Everything humans will ever recognize as an image occupies, provably, essentially none of the space.
Nothing creates images
Now follow the logic where it insists on going. If every possible image already has an address, then no device has ever created one.
A camera is an instrument that looks up addresses. Point it at a wedding and it returns the coordinate of a frame that has been a member of the set since before weddings existed. A painter navigates there slowly, by hand. A sentence typed on a screen is a frame with an address it has held since before language.
And an AI image generator — the technology we argue about most — is not an exception. It is the clearest case. Every image a diffusion model will ever output already had its address before the model was trained, before computers existed. Training does not teach a model to create; it surveys where the islands of meaning lie in the ocean of noise. Generation is navigation. The model's weights — a few gigabytes standing in for a space of 108,426,913 frames — are a compressed atlas of the islands, and the atlas can be that small only because meaning is compressible.
Cameras, painters, writers, and neural networks are the same kind of thing: navigators of a space none of them made.
The only thing a frame can't fake
This is where a decade-old late-night thought lands on the most current problem we have. If the set contains every real photograph, it also contains every convincing forgery — the politician's face in the place they never stood, rendered at every angle, in every light. Not "could contain." Contains. The fakes exist at their addresses right now, waiting to be navigated to, and generators are getting better at the navigation every month.
The consequence is uncomfortable and clarifying: detection is mathematically doomed. No analysis of pixels can separate a captured frame from a navigated one, because they are members of the same set with the same kind of address. A frame proves nothing about itself.
What a frame cannot fake is a witness. A cryptographic signature over a frame's fingerprint, made at a moment in time by a key someone controls, is a fact about the world outside the set: this person vouched for this frame, then. That is why the explorer's registry signs every claim, and why its public archive lives in a repository where the history of who added what, and when, cannot be quietly rewritten. Trust in images was never going to come from the pixels. It comes from provenance — from people signing their witnessing.
A universe you can finish
One resolution in the family is special. At 1×1 the set holds exactly 16,777,216 frames — every possible one-pixel image — and a person can watch all of them in an afternoon. It is the only complete universe of pictures a human will ever finish. All sixteen million frames also fit on a single 4096×4096 sheet, one pixel each: an entire universe rendered as one image. And that atlas is itself a single frame of the 4096×4096 space — one universe carried inside a member of another, the way every 1080×1080 frame is tiled somewhere inside the 1081×1081 set.
That nesting is the last thing the number taught me. Growing the grid by one row and one column multiplies the set by 1015,612, and yet the larger space quietly contains the smaller one's every member. There is no top. Whatever resolution reality is sampled at, its complete set of appearances is a finite integer — unimaginably large, perfectly specified, and already there.
The set contains everything and can testify to nothing. The value of a picture was never its existence — every picture already existed. The value is the address, and the address is exactly as hard to find as the picture. A complete library, whose card catalog is the library.