Sorry if posted in wrong place. Just wanted to share some images i got from qwen ai when fed original low res. Ps1 images. I think in a near future we will have shaders able to remake games on the fly.
Of course this couldnt yet run in real time, but the possibilities with ai is really endless.
Second one is fine, the rest feels a bit off, especially the third one, with the doll face. Whenever AI is used, I think the first step is to nail the style right, not merely upscaling or modernizing.
Where the will exists, there is a way, but compute shaders and pre-calculations / calculating or making pre-rendered graphisc have better chances here. Might be game specific and an upgrade of texture replacement. Ordinary fullscreen shaders alone can’t just be imbued with sufficient IQ like that to handle general situations the way AI upscaling and enhancing are doing it.
I have updated with some more images. I cant believe how good it makes it look sometimes.
My guess is this is why developers pump out remakes lately, since they know it can happen on the fly in a “near” future.
My guess is that we live in an era where creativity is too rare, hence we see remakes or even remakes of remakes, and not only in games, but in movies too.
In topic, I second @guest.r.
Also, such things cannot be called “shaders”, technically; Since the latter always operates on single pixels (a bit of simplification here), while “generative AI” works on the “context” of a whole image.
While this can smell like a niptick statement, it is not, because the bigger the context, the more the generated/shaded content would be bound to it, ofc.
As a result, Working on an image context cannot assure the context would be preserved across frames and as a consequence, “AI” needs to be pre trained on the whole game for that or risk is seeing characters or objects morphing across the game.
Pretraining, on the other side, defeats the purpose of using it like you would do with a shader.
At best, I can imagine better algorithms of edge smoothing/dedithering with less false positives, at least in the near future.
Off topic again, my hope is to see brand new (and I mean it, new) games (IA models certainly would only make it worse).
Sadly it’s not that human creativity is too rare nowadays. There’s still plenty of it all around. But it’s just not cost-effective enough.
But I admit I’m a complete doomer. And if I had to play a game hallucinated like that, then I’d rather go back to smoking weed like twenty years ago. XD
Oh that’s an interesting POV and I hope you’re right.
That would be a retroachivement for sure!
I definitely think this will be possible to use as some kind of filter in the future. Not necessarily called a shader. Perhaps the emulator or whatever it will be will be able to train the filter/shader when loading the game. Of course we probably talk about tens of times faster processors than what we have today. But ai is definitely a huge part of the future in games and everything else. The current issues with it looking wierd and obscure in certain moments is just “new tech immaturity”. It will go away.
No, I dont work with skynet 
i just have high hopes for the future, unless nuclear war unleashes.
