Why Era Does Not Matter to Geometry
A neural network that extracts facial embeddings does not notice when a photograph was taken. It extracts the same structural features from a 1950s film still as from a modern smartphone selfie. The bone structure beneath the face, inter-ocular distance, jaw angle, cheekbone prominence, is the same regardless of whether the image was captured on nitrate film or a modern CMOS sensor. Temporal distance is simply not a variable the embedding encodes.
This means that if your facial geometry is closest to that of a star who has been famous since the 1970s, Ollie will return that match regardless of the decades between you. The geometry matches; the chronology is irrelevant. Facial similarity is about three-dimensional structure, not photographic era.
Why Long Careers Show Up in Your Results
Ollie's database holds living celebrities only, but many of them have been famous for decades. Actors, musicians and public figures whose careers began in the 1960s, 70s or 80s sit alongside today's stars, so a match can point to someone your parents or grandparents would recognise.
Each celebrity is represented by several verified photos, and the network judges them by bone structure. A photo from 1985 and one from last year land in the same region of the embedding space, so the age of a photo does not push a celebrity up or down your list.
What an Era Match Reveals
Receiving a strong match with a celebrity from an earlier era is a signal about facial archetype, not coincidence. Every generation of stars includes particular geometric types, and those proportions transcend decades. If those proportions are yours, the match will appear in your results with high confidence.
Think of the era match as an aesthetic anchor. It identifies which geometric archetype your face most closely resembles. That archetype has been prominent throughout human history and continues to appear in contemporary celebrity faces, but if the clearest example in the database became famous in the 1970s, that is who you will see.