How the 3D model, point cloud, image and video on this page load.
3D models
Most models here come from photogrammetry: overlapping photos solved into a textured mesh. The browser gets GLB files, rendered posters, lighter and fuller model choices, saved camera framing, and source-format notes where they matter.
Texture compression is treated as real delivery proof, not a label. If a model has ordinary WebP textures, it is described that way until an actual KTX2/BasisU asset exists and is served correctly.
glTF and GLB · KTX2 textures
Point clouds
A point cloud is a 3D scene made from many individual points, often with colour, rather than a solid surface. It is closer to a floating swarm of measured dots than a normal 3D model.
PLY is one common source file format for that kind of data. Public viewing works better when the cloud is converted into streamable tiles, because the browser can load nearby detail instead of swallowing one huge raw file.
PLY file format · Point Cloud Library
Photos
Photos are not just one resized JPEG. A good source can become AVIF, WebP, and JPEG fallbacks across several widths, with a tiny placeholder so the layout is stable before the real image arrives.
Some images also keep richer variants such as wide-gamut or deep-colour AVIF. Those stay separate from the everyday fallback path, because a phone screen, a desktop display, and an old browser do not deserve the same file.
AVIF · WebP
Video
Video keeps a plain MP4 path because that is still the boring thing that works. Some posts also ship a DASH package, but it stays behind the media settings player switch instead of becoming the default public browse path.
The page treats a video poster as browse material and the player as an inspect step. That keeps a grid of clips from behaving like every card secretly wants to download a movie, while leaving room for adaptive segment playback when you ask for it.
HLS · MPEG-DASH
Why the variants exist
Browsers, screens, networks, and media types do not agree on one perfect file. The page stays simple, while the saved asset details keep source kind, quality level, preview confidence, dimensions, format, and fallback intent.
Each asset gets the special handling it needs. The goal is not to show off file formats; it is to make the page choose the right thing quietly, then explain the machinery only when you ask.
