Can AI manage a community's shared photo archive?

Hypothesis

Every community produces thousands of photos and no archive — they rot in phone rolls and chat threads. An AI custody pipeline should be able to collect, deduplicate, curate, and file a community's photos with clear ownership and consent boundaries, without a human librarian.

What we built

A photo custody pipeline for one of the self-organised communities we coordinate: submissions flow into a managed archive where AI handles the librarian work — deduplication, quality filtering, categorisation — and the community keeps a single canonical, browsable collection instead of a thousand scattered copies.

Learnings

• Custody is a consent problem before it's a storage problem: who submitted what, and what the community agreed the archive is for, must be decided before the first photo is processed.
• AI curation is good enough to replace the volunteer nobody wants to be — the librarian role, not the photographer.
• Communities trust an archive they can see the rules of; the pipeline's filing logic is documented where members can read it.

Log

• 2026-07-06 — Pipeline live; AI curation running on new submissions.
• 2026-06-28 — Archive structure and consent scope agreed with the community.