What is Landmark Anchoring?

Because AI image models understand only relative magnitude and never actual measurements, they need something concrete to size a product against. Landmark anchoring supplies that by describing the product's position and scale relative to a fixed landmark, on the body or on the product itself, rather than a vague or floating comparison object. Glasses get described as sitting low on the face, the bottom of the lens reaching mid-nose, the top rim clearly below the eyebrows, with a visible strip of skin between brow and frame. A bottle gets described as no taller than the books beside it.

Measured testing on this technique overturned an earlier assumption that hands, faces, and plates all made reliable anchors. A hand alone turned out to be one of the worst anchors tested, with error close to forty percent, and plural anchors, a row of books rather than one book, failed nearly as badly. What actually worked was a single, singular anchor paired with an explicit landmark phrase: one and a half times the cork landed within two percent, and rim level with the seat landed within seven to nine percent.

The rule that follows is specific: pick exactly one clean anchor, tie it to a landmark phrase rather than a floating comparison, and drop hands and plural objects from the anchor list entirely.