After zoom-in, the viewport must keep what the user was attending to
Aliases: zoom focus · lost subject · viewport drift
What it is
A spread is for seeing more clearly something already being looked at: a face, a line of type, an intersection. After the zoom, that thing still has to be in the viewport. If scale goes up and the face is pushed off-screen, people must pan to find it, and zoom-in became two tasks. Keeping the attended object is stricter than pinning the geometric midpoint: the midpoint belongs to the fingers, the attended point to the eyes.
Why it happens
As scale rises the viewport narrows, and any anchor error throws edge objects off screen. People often put a finger on each side of the subject, so the midpoint happens to be the focus and a midpoint anchor is enough. They also put one finger on the subject and spread the other into empty space; the midpoint then biases toward empty space and the subject slides out. Worse, a lift animation recenters the viewport on the whole image and clears the just-enlarged patch. If pan is forbidden during the spread, people cannot steer the subject back as scale climbs. The attended point can be proxied by the contact midpoint or by the pre-zoom viewport center; the wrong proxy loses focus.
Studying it
Name the subject first (eye tracking or “look at this face”), then have people spread. Measure the face’s visible-area fraction, whether it left the frame entirely, and time to a repair pan. Compare anchoring at the two-finger midpoint, at the pre-zoom viewport center, and at a labeled point of interest. Asking “can you still see what you wanted” at the end tracks this claim better than asking whether scale was accurate.
Where it stops holding
When the whole picture is the subject (composition), cropping the edges on zoom-in is expected; not every pixel must stay. A group photo with many faces has no single focus; midpoint is a fair default. Tiny scale steps (105%) almost never throw content out. A following map that locks the viewport to a GPS dot will yank the user back when they spread to look at a nearby building; that lost focus is an external lock, not the anchor formula.
Applying it
- Allow pan during the spread, and do not reset the viewport to the whole-image center on lift.
- If a more credible focus exists (a selection, the last tapped annotation, a visible POI), let it override a purely geometric midpoint.
- Have people point at the glasses on a face, then spread. The glasses should still be on screen and larger. If the first sentence is “where did the face go,” inspect the lift animation and the anchor proxy.