A single photo has no depth — until this model gives it one. With a depth value per pixel we can
fake a viewpoint change: move your pointer over the image and near things slide further
than far things, the way they would if you leaned around them. That's the "3D photo" effect, from
one ordinary picture.
backend –depth latency –warp –
Each frame, every output pixel samples the photo shifted by (depth − ½) × tilt —
a backward warp, so near pixels (high depth) move most. It runs on the main thread only while you
move; on prefers-reduced-motion it holds a gentle static offset instead of animating.
Reduced-motion is on: showing a fixed tilt rather than following your pointer.
