Two people opening the same service see different images attached to the same film. The artwork is not a fixed property of the title but a variable the platform adjusts.
The image is treated as a testable asset
Platforms commission or extract numerous images per title, each emphasising a different element: a performer, a relationship, an action, a tone.
These are then tried against real audiences, and the system records which produced a click and, importantly, which produced a click followed by sustained viewing.
Nothing about this requires new footage, since most images are frames pulled from the material itself and treated afterwards.
Assignment is personal rather than universal
A viewer whose history contains a great deal of one performer's work is shown the image featuring that performer.
Someone who watches mostly comedy sees the frame that reads as comic, and someone who watches thrillers sees the same title presented as tense.
The title has not changed. The argument being made for it has been selected from a set of prepared arguments.
The optimisation target matters
Optimising for clicks alone produces artwork that overstates what a title contains, which generates disappointment and abandoned viewing.
Platforms therefore weight for viewing that continues, since a click followed by an exit within minutes is worse than no click at all.
That constraint limits the exaggeration but does not remove it, because the emphasis can still be misleading while remaining accurate.
Creators lose control of the first impression
A director who composed a specific image to represent a film may find it replaced by a frame chosen for measured performance.
Minor characters are sometimes given prominence because their presence tests well with a particular group of viewers.
The objection is not vanity. The image is the first statement a work makes, and having it selected by a system optimising a different objective changes what the work appears to be.
Why rows and ordering work the same way
The same approach governs the category headings on a home screen, which are assembled per viewer rather than fixed.
Both the contents of a row and its position are decided by the same estimate of what a particular person will engage with.
The interface therefore presents itself as a curated shelf while functioning as a continuously adjusted argument aimed at one household.