How the pixels get here

The page starts with the smallest useful image, then upgrades to sharper AVIF, WebP, or JPEG variants when the browser and screen can use them.

The fallback stays ordinary on purpose: older browsers still get a real image, and better displays can ask for wider-gamut or experimental HDR files when those exist.

A normal photo here is not one file. It is a little stack of AVIF, WebP, and JPEG versions at different sizes. The browser gets to pick what fits the screen.

The boring JPEG fallback is not shameful. It is the emergency exit. A site without an emergency exit is just a confidence trick with pretty typography.

"HDR" means high dynamic range: more room for bright highlights and dark detail than a normal SDR image. It is a display and encoding promise, not a magic quality button.

The normal photo path still uses SDR AVIF, WebP, and JPEG. Some posts also expose opt-in Display P3 AVIF files in Nerd facts: ordinary 10-bit PQ HDR files ending in -nerd-hdr-pq.avif, and AVIF gain-map files ending in -nerd-gain-map.avif.

The gain-map version keeps an SDR-looking base image and carries extra HDR recovery data for software that understands it. If the browser, operating system, or screen ignores the extra data, the ordinary SDR path is still the safe view.

HDR is not a button marked Better. It means the file and display can talk about brightness beyond the usual SDR range. If any part of that chain lies, the result can look expensive and wrong.

For now, HDR photo files live behind Nerd facts. That keeps the public default sane while the experimental path can still be poked, compared, and blamed properly.

The standard path is MP4 because it works almost everywhere. Some posts also have DASH for adaptive playback, controlled from Media settings.

A poster keeps browsing light. Opening the video is the signal to spend more bytes.

Video is where ambition becomes expensive very quickly. A good default starts fast, plays widely, and does not make a phone download a giant prestige file just to see whether anything interesting happens.

DASH exists for adaptive playback, but it is still treated as a deliberate detail path rather than something every casual visitor should pay for immediately.

A panorama begins as a wide 2:1 image. The viewer wraps it around you so dragging feels like turning inside the scene.

Cards use a poster first. The full viewer loads after intent, with drag controls, fullscreen, and the best published sphere source.

A 360 view is usually an equirectangular image. It looks weird as a flat file, then makes sense when a viewer wraps it around the camera.

The card should stay light. The immersive viewer loads after intent because dragging around a sphere is fun, but it is not free.

Most 3D views come from photogrammetry: photos turned into a textured mesh. The page starts from a saved camera angle so the object opens in a readable pose.

Lighter and sharper model rungs are separate. Texture labels only appear when that format is actually published.

The 3D models mostly come from photogrammetry: lots of photos, a reconstruction pass, a mesh, textures, and then a public GLB that a browser can survive.

Model files can get heavy fast, so the site keeps separate rungs for first look, standard view, and source-adjacent proof when that exists.

Map views are stitched top-down images. Large ones are served as tiles so zooming in asks for detail only where you are looking.

That is why a map can open quickly and still have sharper pieces waiting deeper in the zoom.

Map views are stitched top-down images. They are useful because they show the actual captured surface, not a generic map provider's idea of the area.

Large maps use tiles so zooming asks for detail where you are looking. Loading the whole thing upfront would be honest, and mostly useless.

Terrain views use height data to show shape: slopes, bumps, gaps, and the places where reconstruction gets suspicious.

When a richer terrain package exists, it streams as small 3D pieces. It is useful for inspection, not a claim that the scan is survey-grade.

Height views are for reading slopes, bumps, and reconstruction weirdness. They help explain shape where a flat image only shows color.

If the data is imperfect, the interface should stay honest. A nice shaded surface is not permission to pretend the capture was survey-grade.

A point cloud is the scene as many colored points instead of a solid mesh. It is good for seeing what the reconstruction actually measured.

Public clouds use streamable tiles when available, so the browser loads the part you are looking at instead of one huge source file.

Point clouds show the measured scene as many colored points. They can look raw because they are rawer. That is often the point.

The public path uses streamable 3D tiles where possible, because throwing a dense raw cloud straight at a browser is how you turn curiosity into a frozen tab.

Deep zoom is for giant still images. The image is cut into tiles, and the viewer loads the visible pieces at the zoom level you need.

That keeps huge stitches inspectable without making normal browsing pay the full cost.

Deep zoom is for images too large to treat like normal photos. The viewer loads tiles at the zoom level and position you are actually inspecting.

This is not glamorous. It is just the difference between an inspectable huge stitch and a browser tab giving up.