Yes, but it has to make up details to do that because it gets a lower resolution image, or for frame gen it gets two frames it needs to interpolate

if I understand right, DLSS (outside frame generation, a relatively newer DLSS feature) isn't interpolating frames, it's upscaling frames. and its never before altered the look beyond just upscaling it.

this new shit is definitely a different direction for DLSS and it sucks and I hate it

You can write a shader that drastically alters the artwork too, and DLSS is basically a closed source ML shader running on a separate chip (come at me gamedevs 😤)

edit: tho don't get me wrong it does suck. I don't know why would I ever want a face filter in a video game

There has been machine learning for decades, but you're right in that interpolating frames is quite different from the way generative AI works. Both DLSS versions rely on a form of neural network but the latest ones are diffusion models which are more controversial, for reasons I'm sure you know.