It also really bothers me that DF already talk about how more accurate the lighting of the model is. It might overall feel like it is to them, but in the few examples they showed, there was already one scene, where a strong directional spotlight on a face got replaced by a softer frontal diffuse.

The first thing the model does is Yassify the scared 20 year old like the old Aloy meme.

Still, DLSS kinda was the single consumer focused genAI product I think you couldn't really dismiss as bad, useless or amoral. It's just pretty damn good. But the next version seems to just create an uncanny version of everything as seen through a an AI lens.

It should be noted that both this, as well as the upscaling and frame gen aspect of ARE in fact genAI. They might not be prompt based, but instead use all the normal TAA inputs, but the network still generates completely new images, where you can't correlate pixels from the output to the input.