It is implemented using web maps technology (https://leafletjs.com/), similar to e.g. the Google Maps satellite view. The screenshots are then served pre-assembled into quadratic map tiles at different zoom levels. This way the client only ever has to load and display a hand ful of relevant tiles.

It's pretty clever optimization, it actually only shows a few dozens images at all times and only update the groups as you zoom in, for example https://tiles.onemillionscreenshots.com/2024-05-01/tiles/-1/...

You don't even need something fancy like WebGPU/WebGL, just the CSS transition property will do the job there.

I assume that they are using some form of mipmapping, a technique by which you have several images at different sizes to represent different levels of detail. It is used a lot in 3D environments.

https://en.wikipedia.org/wiki/Mipmap