X generates a set of resized copies of every uploaded image and picks one to display depending on context. The timeline uses a small variant so feeds scroll quickly; opening a post loads something larger. None of these is the file that was actually submitted. When you right-click and save, the browser writes out whatever it had already loaded, which is a compressed preview.
For viewing on a phone that is completely fine. It stops being fine the moment you enlarge, crop or print, because you are working from a copy that was already reduced twice — once by X on upload, and again by the variant your browser happened to fetch.
Requesting the original variant skips the second reduction entirely. Nothing is processed or enhanced; it is a request for a file that was already sitting there at a different address. The difference is usually obvious side by side on anything with fine detail — text in a screenshot, foliage, faces at a distance.
Worth being honest about the ceiling, though. X re-encodes on upload and caps stored dimensions, so the original variant is not the photographer's raw file. A 6000-pixel photograph comes back around 2048 on the long edge. What you get is the best X holds, which is a meaningful improvement over the preview and still not the source.