An individual has developed an automated pipeline to streamline the time-consuming process of scanning and editing 35mm film, moving from a manual four-hour workflow per roll to a highly automated digital lab.
Automating Film Scanning and Editing with AI and Local Models
The initial attempt to use Claude to control the scanner via SANE (Scanner Access Now Easy) failed after the agent incorrectly managed the scanhead direction, requiring physical repairs to the hardware. Following this, the developer transitioned to using VueScan for scanning and built a customized software suite for post-processing.
The automated workflow utilizes a combination of specialized tools and AI models:
- Image Conversion and Color Correction: A custom Python routine using NumPy handles negative inversion and color balancing, using the blank leading slot of the film as a reference for consistency across frames.
- Dust Removal: To handle imperfections, the system uses image analysis to detect dust. It employs LaMa, an inpainting model, to fill detected spots by copying texture from nearby film areas to ensure natural-looking repairs.
- Automated Tagging: Once processed, Qwen3-VL runs locally via Ollama to inspect finished frames and automatically generate descriptive tags, creating a searchable digital library.
To share the results, the developer built a web gallery hosted on Cloudflare R2, featuring an interactive 3D interface built with three.js that simulates picking up film canisters from a felt table to view the images.
Sources
- Automating my 35mm film scanning pipeline （Hacker News Frontpage, 2026-10-03）