AI Photo Management Guides

Overview

Two capabilities do almost all the useful work: matching a plain-English description against the content of an image, and grouping faces so you can name a person once instead of tagging them ten thousand times. Both are mature, both are unglamorous under the hood, and both run acceptably on an ordinary CPU, which is why Pluto Photos ships [contextual search](/features/ai-contextual-search/) and [face detection](/features/ai-face-detection/) as models that run on your own machine rather than a call to a cloud API.

The important structural fact is that the expensive part happens once. Indexing runs a model over each photo a single time and stores a few hundred numbers; searching afterwards is arithmetic over those numbers and costs almost nothing. That asymmetry is exactly why this does not need a data centre, why a subscription is not technically necessary for it, and why both features work with no internet connection at all once a library is indexed.

It is worth being equally clear about the limits. These models are good at scenes, objects, settings and general activities, and bad at proper nouns, specific places and anything to do with time. They rank by similarity rather than matching a condition, so there is no such thing as an empty result, the bottom of any result list is noise. The guides below go into both sides.