Schaller says Claude Code drafted a SANE backend for his film scanner

Christian Fredrik Kalager Schaller posted Using Artificial Intelligence to reverse engineer hardware on 15 October 2026. The blog’s site description says he is writing as a member of the open source community, and that official Red Hat communication is on Redhat.com. Phoronix, Michael Larabel, 15 October 2026 at 10:29 AM EDT, calls him Christian Schaller of Red Hat. The post is the personal blog.

The scanner he bought

Schaller writes that he bought a Plustek OpticFilm 8200i expecting Linux support, then found the chipset had changed. He names a Genesys GL128 controller, USB id 07b3:1825. He says an older scanner with the same product name uses a GL845, which SANE already handles.

He writes that an earlier attempt with Claude Code stalled for months. The method he describes next comes from a post by Tom Crider, GloriousEggroll, which he says showed an agent with a Windows instance capturing what a vendor driver does. Schaller says he pasted that post into Claude and asked for the same kind of capture for the scanner: the vendor driver in a Windows virtual machine, USB traffic recorded on the Linux host, then a comparison with captures from his own driver. He says that produced a first recognizable scan within a few hours, with quality problems, and a few days later a backend he calls near complete, including infrared.

The linked post is dated 26 September 2026. Its text describes a Windows session, driven from Linux, used to gather device data for Linux drivers. It does not name a mouse. The mouse-driver description is Schaller’s account of the post.

He is explicit that hw-harness is a later generic form of the scripts he used, and that he did not run these projects on the published harness. He says the person holding the hardware still decides what to build and checks that it works.

What he says the backend does

He says the backend covers color, gray, infrared, and color plus infrared, resolutions from 300 to 7200 dpi at 8 or 16 bits, infrared dust and scratch removal, and an exposure multiplier. He packages it as a Fedora RPM. He says he mostly tested it with Negative Vibes, a Rust GNOME app he also had Claude write, for color and black-and-white negatives and slides. He says that app is tested on the 8200i so far, and that SANE still has no standard way to expose an infrared channel, which the app would need for other film scanners.

He then writes that upstream SANE’s genesys backend gained support for this scanner in September 2026, while his own attempt was paused. His comparison page labels a full-frame pair at 3600 dpi as 34.6 seconds for his backend and 54.4 seconds for genesys, says both resolve the same fine detail, says his backend has the infrared channel and genesys does not, and says the frame was scanned on 9 October 2026 on an OpticFilm 8200i. He calls the comparison one scanner and one frame. The repository is cschalle/OpticFilm8200i-sane. He says he published it as a proof of concept and expects to use the upstream driver once it matches his.

The same day, at 15:10 UTC, he opened sane-project/backends work item 886, still open when read. The title is “Alternative SANE driver for Genesys GL128 (Opticfilm 8200i)”. In it he says upstream merged support about two weeks earlier. The blog’s “September 2026” and the issue’s “two weeks ago” are both his wording. The upstream commit date was not opened for this story. He asks whether a patch from his backend would be welcome. That is not an upstreamed driver.

Phoronix spells the scanner Pustek, calls the result a basic driver that is now working, and does not mention the upstream genesys support. This account follows Schaller’s spelling and his caveat.

The laptop camera, in his words

The same post describes the MIPI camera on his Lenovo ThinkPad X1 Carbon Gen 14 (Aura Edition): a Sony IMX471 behind an Intel IPU7. He says Linux images were dark, vignetted, and yellow-green, and that one day’s session produced kernel and libcamera patches. He says the kernel patches have gone to Kate Hsuan, and that Wim Taymans is reviewing the libcamera patches. A low-light test he reports moved frame brightness from 31.1 to 90.9. He writes that the scans look decent and the camera looks better, and that he is not claiming either is finished. Those results are his. Notes and files are on mipi-fixes-x1-aura.

A separate Claude Code thread in the kernel is Linus Torvalds’s page-cache branch, covered in the 12 October account. That branch names Claude Code in the commit. It is a different change, with its own measurements.