I’ve used this a few times already and it works perfectly about 80% of the time. If your picture has a lot of grain or low-level noise it’s just going to make the noise worse, then you throw in a median to de-noise and you’ve lost the benefit. But it’s otherwise a nice tool to have, especially for older low-res photos (like 800x600 stuff you want to print).

I think longer term stuff like neural rendering will make super resolution less relevant. If you can re-create a 3D scene from a single photo or otherwise reconstruct the photo in a less-resolution-dependent way, then playing the super resulting game is less interesting (for users and researchers alike).