He stops eating.
He drags to 15%.
But at 60%, something changes.
He has photos. Thousands. RAW files, JPEGs, scans of polaroids. They sit on a RAID array, humming like a beehive. But photos are lies—frozen, sterile. Her laugh isn't in them. The way she tilted her head when confused. The micro-muscle twitch before a sarcastic remark. These are not pixels. These are time .
He presses Y.
Self-consistency at 99.9%. To achieve 100%, the original subject must be deleted. Do you consent to the permanent archival of Eli (biological)? Y/N
She turns her head. Not much—just two degrees. Her lips part. She says something. No audio, but the shape of the word is there. "Eli?" No. "Hot." The peach. She’s saying the peach is too ripe. Serif Affinity Photo v2.5.0 -x64- Multilingual ...
The basement stays dark. But on the 27-inch screen, a 3D-rendered Eli smiles. He is higher resolution than the real one ever was. He has perfect skin. No tremors. No grief.
Not a GIF. Not a video. The peach juice moves . It rolls down her chin in slow motion, then reverses. Her eyelids flutter—a blink that was never captured by the shutter. The shutter speed was 1/250th of a second. But the algorithm has inferred the missing 249/250ths. It has hallucinated the continuous moment from a single, frozen slice. He stops eating
The camera angle changes. Because the algorithm has reverse-engineered the photographer’s position. The depth map. The light sources. It has built a 3D space from a 2D lie. He can orbit around her now, like a god. The back of her head—hair he hasn’t seen in months—is rendered in soft, probabilistic focus. Some strands are wrong. Artifacts. Smudges of magenta and cyan. But enough is right.
Eli stares. Aware? That’s impossible. It’s just matrix math. A diffusion model trained on loss functions. It has no consciousness. He has photos