Clean-room capture target
Every word on this side of the divider is rendered as plain HTML in a default system serif. There are no images, no web fonts, and no third-party URLs anywhere on this page. That is deliberate: with zero cross-origin assets, the serialized image can never taint the canvas, and the texture upload can never raise a SecurityError.
A serif face at high contrast gives the magnifier something honest to chew on. The thin and thick strokes, the bracketed feet, and the sharp terminals all carry fine detail that survives serialization and reveals exactly how cleanly the captured texture is being sampled on the other side. Near-black ink sits on a flat white field so the lens has the widest possible tonal range to enlarge.
The browser paints this column. A deep clone of it is serialized into an SVG foreignObject, rasterized through a native image, and handed to the graphics context as a single texture bound to a sampler. A small fragment program then reads the texture per pixel, and inside a circular region it pulls the sample coordinates toward the lens center, enlarging whatever sits beneath it. Select any text here and the highlight shows up in the magnified copy too.
Hover the panel on the right and drag the lens across the magnified copy. The text you are reading right now is the same text being enlarged there, pixel for pixel, with no smoothing tricks beyond a linear texture filter. If the right panel shows these sentences blown up, the pipeline works end to end.
Same origin only. System serif only. No network. No taint.