What if you can make concessions to make the game look this good, without it actually looking this good. Like checkerboarding looking almost like 4K but being way cheaper. Maybe real time rendering is more about efficiency and speed that just letting render farms render them. These two disciplines are completely different.
Let's say the woman in this photo is from a real time cutscene on PS5. With the SSD and
advanced culling solutions (there was a crazy microsoft tech demo on this), it's not impossible to have this level of detail when the camera is close to her since the PS5 will only have to bother rendering the face while lowering the resolution of background objects with
VRS because its hidden behind a shallow DOF. As the camera backs off the
geometry engine will lower the polygon count while being imperceptible to us. At the same time the SSD will hammer new objects into the scene that didn't need to be there 1 second ago. Other things like
virtual texturing will make this transition even easier to achieve.