Automating my 35mm film scanning pipeline

A photographer recounts rebuilding a 35mm film scanning workflow to spend less time processing and more time shooting. The original routine - lab development, cutting negatives, manual scanning, inverting and editing in darktable - was taking about four hours per 36-frame roll. An early experiment to fully automate scanning with a Claude agent driving SANE backfired, mechanically stressing the scanner when scanbed directions were inverted, so the decision shifted to keep VueScan for capture and automate the downstream steps instead.

The resulting pipeline automates negative-to-positive conversion, dust removal, metadata, and publishing. Negatives are inverted in film density with a numpy routine that normalizes color across a roll using the blank leading slot as a reference. Dust detection uses context-aware image analysis to avoid mistaking sun glints for defects, with LaMa inpainting to repair marked spots and thresholds for human review on grainy areas; per-frame exposure, contrast, and temperature controls remain available. Qwen3-VL run locally via Ollama drafts tags and combines camera, lens, and film data into a searchable library. Picked images can be published to a Cloudflare R2 bucket and showcased in a playful 3D film-canister gallery. The build took 11 days and 64 commits and now serves 22 rolls (314 photos), reclaiming creative time while keeping hands-on photography enjoyable.