Yannick Richter bought a second-hand Epson V370 flatbed scanner. He didn't plug it into a computer. He didn't open the scanning software. He took it apart.

Inside he found a Sony ILX561K linear CCD array: twelve rows of pixels, 122,400 light-sensing elements arranged in a fine line. This sensor had been reading light for years — light bounced off documents through a glass plate. Richter's question was different: what if that light came through a lens instead? He spent roughly a year reverse-engineering the sensor's timing protocol, redesigning the signal capture circuit, mounting the sensor on a 100mm precision linear drive controlled by a Raspberry Pi 5 and RP2350, and positioning a Pentax medium-format lens in front. The mechanism that once moved across flat documents to scan them became a shutter that moves the sensor across the focal plane to expose it. The result: images at 80,000 × 40,160 pixels — 3.2 gigapixels, 16-bit true color.

The 余项之美 of Project Gigapixel isn't primarily the resolution. Resolution is a symptom. The deeper structure is this: the flatbed scanner was born in the late 1980s and has existed as a "document reproduction apparatus" for over 35 years. Its product definition — faithfully convert flat objects to pixels — hardened into a complete already-construct. Every sensor specification, calibration routine, and driver exists to serve this definition. Inside that definition, the linear CCD architecture is an engineering choice, not a photographic proposition; exposure time is a calibration parameter, not a shutter; the requirement for stillness is a suppressed technical precondition, not an aesthetic position. Richter's chisel was a single act of disassembly: removing the sensor from its document-reproduction mission and letting it face a lens. Once that chisel-construct cycle completes, a double remainder becomes visible.

The first remainder: the imaging capability of the linear CCD, which was present from the moment the chip was made. The construct "document scanner" suppressed it by definition, but it never went away. The second remainder is stranger and deeper: the temporality of scanning as a photographic fact. The linear sensor must traverse the focal plane over several seconds to minutes to complete one exposure. This means the camera can only photograph what doesn't move — subjects in motion disappear mid-scan; only what holds still can be recorded. This is not a limitation. It is a photographic proposition: the camera doesn't ask "what is interesting?" It asks "what is willing to wait?" Richter photographed instruments on a workbench; cropped in, you can read the value printed on a 0603 SMD resistor.

Why see this now rather than later? In 2026, AI image synthesis can produce "a high-resolution photograph of anything." But Richter's camera can only capture what actually exists and actually holds still — it photographs the world's patience. No AI can simulate that temporal commitment. The naming gap remains wide open: this is not scanography (which places objects directly on flatbed glass without a lens — a completely different practice); not gigapixel stitching (multiple normal shots merged); not DIY camera building (that label hides the structural proposition). Richter presented at Hackaday Europe in July 2026. There were no photography critics, no art market players — only engineers and hardware enthusiasts. That is the characteristic habitat of a remainder: it lives most vividly where it has no name.

Han Qin (秦汉) · @hqinjarsy · SAE Judgment & Aesthetics