First off this is a video and not capture of a remastered version of Blizzard's iconic Diablo II and Diablo II: Lord of Destruction expansion. Instead this is the result of Deep Learning Neural Network AI doing it's magic over the course of 10 days and nights on some video. Taking the 800x600 resolution of the original (and its 25fps) up to 4K 60.
The results are impressive to say the least. Check it out.
The work of YouTuber Michael ReznoR, he uses AI to not only upscale but enhance and smooth-out that aRPG classic. He notes that it in addition to taking over a week to enhance the footage, 50244 files and 390 GB of disk space were used. So yeah, a lot more than applying a filter.
It's been rumoured for some time that Blizzard is working on a remastered version of Diablo II - ala StarCraft Remastered, which took the original sprites and recreated them in HD. But, recreating and redrawing all of Diablo II's assets would be quite the undertaking - with several reports form those that worked on its initial release pointing to this potentially being a painstaking and difficult process due to how the original assets were stored and created.
But, if the results of AI can look this good - perhaps this is the way to go when re-releasing classic games from the when 640x480 was considered high-res.
Yeh, I am wondering if it can get to the point where old assets can be more than just upscaled nicely, but can have new details added to them. Without knowing much about it, I imagine it's easier or at least neater for 3D games because it could dynamically and iteratively update the models, whereas sprites it would need to update each frame in a way that was consistent.