A Linux user documented how artificial intelligence helped reverse engineer proprietary hardware components, resulting in functional drivers for a film scanner and a laptop camera.

The project began when the individual inherited a collection of undeveloped film negatives and sought to digitize them. After purchasing a Plustek 8200i scanner that initially lacked Linux support due to a chipset change, they explored AI tools to overcome the obstacle. Using the Claude AI platform, they configured a Windows virtual machine to capture USB traffic from the scanner, enabling driver development.

Within hours, a basic scan function operated under Linux, producing recognizable images despite quality limitations. Subsequent refinements led to a near-complete driver supporting infrared scanning and dust removal. Notably, the upstream SANE project later incorporated support for the same scanner model, rendering the custom driver partially redundant.

The same methodology was applied to address issues with the MIPI camera on a Lenovo ThinkPad X1 Carbon Gen 14. The camera suffered from poor low-light performance, vignetting, and color imbalance due to driver defects. By analyzing USB traffic comparisons between Windows and Linux implementations, the user identified and corrected three specific kernel-level bugs: an incorrect crop register declaration, improper width specification in a test register, and insufficient analog gain limits.

In the libcamera software stack, missing lens shading correction and flawed white balance calculations were also rectified. Testing showed frame brightness increasing from 31.1 to 90.9 in low-light conditions, with color ratios approaching those of a standard USB webcam. Kernel and libcamera patches have been submitted to maintainers for integration.

The user emphasizes that AI handled repetitive tasks such as traffic analysis and probe generation, while human oversight remained essential for hardware validation. They released a genericized version of their usbmon capture and comparison framework, termed hw-harness, to assist others in similar reverse engineering efforts.

The account illustrates a practical application of AI in hardware development, where automated analysis reduces the time-intensive aspects of driver creation without replacing hands-on testing or domain expertise.

---

Project 1: A SANE backend for the Plustek OpticFilm 8200i scanner enabled color, grayscale, infrared, and combined infrared-color scanning at resolutions from 300 to 7200 dpi. It supported 8- and 16-bit channels, included infrared dust removal, and offered an exposure multiplier for dense negatives. Distributed as a Fedora RPM, it outperformed the upstream driver in speed at full resolution (34.6 seconds vs. 54.4 seconds) but lacked some resolution steps. The upstream project later added support, making this implementation a transitional solution.

Project 2: Negative Vibes is a GNOME application written in Rust that processes scans from compatible SANE scanners. It preserves raw 16-bit linear data with infrared channels, applies non-destructive adjustments, and supports IT8 calibration via ArgyllCMS. Output formats include TIFF, JPEG, PNG, and DNG with embedded ICC profiles. Available as a Flatpak, it currently supports the Plustek 8200i and aims to standardize infrared channel exposure across film scanners.

Project 3: Fixes for the MIPI camera on the ThinkPad X1 Carbon Gen 14 addressed sensor-level issues in the Intel IPU7 driver and libcamera ISP. Corrections included proper crop rectangle handling, accurate test-pattern register width, improved analog gain limits, and restored lens shading correction. These adjustments enhanced low-light sensitivity by nearly three stops and improved color accuracy, though blue channel performance remains slightly subdued in dim conditions.

The author notes that while neither the scanner nor camera is perfect, the improvements represent meaningful progress. All claims were validated through direct hardware testing, and the shared framework aims to lower barriers for others undertaking similar reverse engineering work.