AI-based upscaling for old games

[ttsiodras]

With all the excitement and hype surrounding AI, one thing that I haven't seen... is the premise of using AI upscaling for old games.

What do you guys think? Would it make sense to use AI algorithms to upscale the large background images of LucasArts / Sierra games?
Or would it be a violation of the artists' original creations?

To have a sense of what it could look like, I used an online service to do a short experiment... and uploaded the resulting video as a standalone resource on my blog.
Here it is - before-and-after from LucasArts' "The Dig"... covered with music from the game:

https://www.thanassis.space/TheDig_AI.mp4

[GateKeeper]

With all the excitement and hype surrounding AI, one thing that I haven't seen... is the premise of using AI upscaling for old games.

You haven't been following the "scene". There have been lots of discussions about that during the past five years on many forums and discussion channels. Here's the thread about the subject on this forum:

viewtopic.php?t=14787

What do you guys think? Would it make sense to use AI algorithms to upscale the large background images of LucasArts / Sierra games?
Or would it be a violation of the artists' original creations?

It can be done, obviously, but it makes all those old games look very cheap.
You could upgrade every painting in the Louvre by turning them into holograms, but very few people would think that's a good idea.

I personally feel that those old games should be played as is, but others have different opinions about that. I'm sure all this is something for the modding community to work with. But when publishers start releasing atrocities, such as "special editons" of Simon the Sorcerer or Blade Runner, that's where we are really crossing the line.

[Praetorian]

I appreciate a few of the proof of concept demos I've seen, but as far as I can tell (or has been divulged by the creators) they all needed a lot of additional work done manually over the result of the AI upscaling.

Because the "lazy" upscaling often produces artifacts, removes details or "misinterprets" a detail for something else entirely and the result is a crispy but worse image, at least for someone familiar with the original in-game image or the original concept art.

I think it might be done well, if dedicated fans put a lot of --probably voluntary, unpaid and largely unappreciated partly due to the use of AI tools-- their own time to do post-editing work on these *and* the AI was trained on original concept art and original high resolution assets (if available) to begin with. The creators would somehow have to ensure that each little detail would not be lost or morphed into something else. Which, I understand, has practically little chance to ever be done at a full game scale.

The other thing is, even if one does find the automated process on its own acceptable, the distribution of the upscaled assets goes against copyright law, unless it's done by the official publisher/ with permission from the IP owner. I guess alternatively, it could be done as a preprocessing step at each user's device (like running a script before playing the game to prepare the assets) or at run-time/real-time while playing the game (which could be very taxing on the system's resources and thus prohibitive).

Plus, you'd have to make sure the game's engine is adjusted to support these new resolutions, which might be another can of worms no one would like to open.

[Red_Breast]

Just to add to GateKeeper's post somebody (Ravippe) made a thread last year.
Ravpippe seemed to be doing something (adding to to Myles Johnston's work?) but I don't know what happened.

viewtopic.php?t=17381