A picture in, a model out
Two ways to get a mesh: describe a shape and take a printable STL away right now, or hand over an image and let a model reconstruct it.
The parametric lane works right now
Pick a shape, set its dimensions, and download a real binary STL. It runs on ordinary CPU with no account, no key and no queue, and what comes out is ready for a slicer.
The image lane needs a GPU
Single-image reconstruction is model inference. This deployment reports honestly whether a worker is reachable, and if none is, the page says so instead of offering a button that fails.
It says what it made
Every model states its source, its method and its poly count read from the exported file. A low-poly single-view guess is never described as a scan.