In an account published October 15, 2026, Christian Fredrik Kalager Schaller said Claude Code helped him build a SANE backend for a Plustek OpticFilm 8200i and address Linux camera defects on a Lenovo ThinkPad X1 Carbon Gen 14 Aura Edition. Schaller chose the work and tested the results on the hardware himself.
From scanner USB traffic to a SANE backend
The scanner case involved a newer Plustek OpticFilm 8200i variant with a Genesys GL128 controller and USB identifier 07b3:1825. An older scanner with the same model name uses a GL845 controller. Schaller said he ran the vendor driver in a Windows virtual machine and captured USB traffic while operating the scanner, then used the captured data in his driver work.
He reported getting a recognizable scan under Linux within hours, followed some days later by a backend with infrared scanning. The backend supports color, grayscale, infrared, and combined color-and-infrared modes, at resolutions from 300 to 7200 dpi and 8- or 16-bit depth per channel. It also supports infrared dust and scratch removal and an exposure multiplier for dense negatives.
Schaller said SANE’s upstream genesys backend had gained support for the newer OpticFilm 8200i in September 2026, before he discovered that work. His separate backend was therefore a parallel project, not evidence that the scanner lacked upstream support.
In Schaller’s single-scanner, single-frame comparison, both backends resolved the same fine detail at 3600 dpi. At full resolution, his backend took 34.6 seconds and the upstream backend 54.4 seconds. That is one comparison under a specific test setup, not a standardized performance benchmark.
What changed in the ThinkPad camera path
The camera case involved a Sony IMX471 sensor behind Intel IPU7 in the Lenovo ThinkPad X1 Carbon Gen 14 Aura Edition. Schaller described Linux images that appeared dark, vignetted, and yellow-green, and changes to both the kernel driver and libcamera’s software image signal processor (ISP).
The kernel work addressed a crop-selection error, an incorrectly declared test-pattern register width, and an analogue-gain cap that Schaller said cost about 1.8 stops of sensitivity. The libcamera work added lens-shading correction and fixed white-balance statistics that mixed the sensor’s native bit depth with an 8-bit black level.
In Schaller’s controlled low-light test, the frame-brightness reading rose from 31.1 to 90.9. The red-to-green ratio shifted from 0.890 to 1.047, and the blue-to-green ratio from 0.672 to 0.823. He said blue remained about 17% low in low light. Those figures describe that test condition.
Schaller said he had submitted the kernel patches to Kate Hsuan. Wim Taymans was reviewing the libcamera changes for submission.
The later harness and the human role
Schaller said the scanner and camera projects used bespoke scripts; he documented the generic hw-harness afterward and did not use it in that later, generalized form for the original work. The harness describes a workflow that records a vendor driver’s activity in a Windows virtual machine, captures USB traffic from the Linux host with usbmon, repeats the operation under Linux, and compares the traces. For devices such as internal MIPI cameras, it proposes Windows-side inspection instead.
The hardware still had to be operated and the results checked in person: Schaller said people decide what to build and test the changes on the physical device.
