Find a 35mm photo of the same station at anything approaching a comparable angle, or ships. Compare this downscaled photo of DS9 next to the 3:10 freeze-frame.
https://imgur.com/gallery/BaaIT/comment/955550109
https://www.christies.com/lot/lot-deep-space-nine-model-4780057/?
Or any still image 35mm photo of the station at close range. The real model is tons more intricate. Topaz AI, et al, still won't pick up all of that. The youtube video is another "more of the same" as this:
https://trekmovie.com/2019/03/15/star-trek-fan-tests-using-ai-upscaling-on-deep-space-nine-clips/
Another attempt at upscaling... Note the starship being blurry as everything.... and the station looks like a big bowl of wax fruit.
It all looks like the usual sort of edge enhancement with overdone edges, mushy innards, overblown halos around edges, very waxy/glossy/plasticky appearance and typical of these techniques. Not anything approaching genuine 4K/5K/whateverK. The level of post-processing compression certainly doesn't help matters. They can remap an image from a lower resolution to a higher one, but adding proportional detail is simply. not. there. Not when going from interlaced 240P (480i), deinterlaced to 480P, then streteched to something like 9x its original deinterlaced size. Or rather, 5K is 14,745,600 pixels combined. 480P is comprised of only 345,600.
Gamut shifting and minimizing bloom or crush is one thing, but from dozens of these sorts of videos the underlying core issue of "Never The Same Color" (or "NTSC" for short, as engineers of the time sardonically called it) still remain present.
All one need to is to find a proper 35mm scanned image, then take the same frame and do the upscale game. Place them side by side and the results will be crystal clear. Or go a step farther and take both frames and downscale them to 480p. Either which way, the result is the same. (Downscaling also strips detail and would make both images look more similar, but telltale signs will still exist.)
Upscaling is nowhere near in the same league as going back to the higher definition source material. We've all seen photos of models, which are models but don't look like waxy cartoons.
