Then the webcam’s tiny LED flickered. Once. Twice. Three times.
She’d bought it for $14 from a surplus bin. The specs were unremarkable: an F/2.0 aperture, a fixed 8mm focal length, and an “8 Driver” architecture that suggested eight parallel imaging pipelines. Cheap. Mass-produced. Perfect for her side project: training an AI to recognize micro-expressions.
She ran a diagnostic. The wasn’t a hardware feature. It was a patch. Someone had written a low-level driver that allowed eight simultaneous video streams, each tuned to a different wavelength. Standard webcams see RGB. This one saw into near-infrared, ultraviolet, and something else—a band the driver labeled SIGMA_8 . Web Camera F 2.0 F4. 8mm-8 Driver
That was six months ago. The day she’d died in a car crash.
She stared at the screen. The camera’s 8mm lens—wide enough to catch a whole room, short enough to distort reality—had recorded her ghost learning to type. Not haunting. Learning. The driver was recycling her last conscious moments, frame by frame, through eight parallel temporal buffers. The camera wasn’t watching her. It was replaying her. Then the webcam’s tiny LED flickered
Morse code: I M H E R E
Elara unplugged the camera.
On frame 12,009, the ghost turned and looked directly into the lens.
Dr. Elara Voss never expected to find a soul inside a driver log. But there it was, buried in line 847 of the firmware for the — a device so generic it had no brand, only a serial number and a prison-gray plastic shell. Three times
Elara patched the feed into her AI. The AI hesitated, then printed: MOTION PATTERN MATCHES 92.7% WITH SUBJECT: ELARA VOSS. TIMESTAMP: 2024-11-15 14:03:22.
She didn’t sleep that night. But she didn’t throw the camera away, either. Some ghosts don’t need a house. They just need an 8mm lens, an f/2.0 aperture, and a driver that remembers them better than any human ever could.