Yep. My favorite image to drive this point home is this one:
We're pushing more pixels, polygons, effects, and geometry than ever before by a wide margin but the more we have the less we notice.
As much as I agree with that the more complex the graphic become the less improvement we see (it starts to be all about the details that make the difference), this picture is not a good comparison. The model used has no good details to show the proper differences that are still between 6k and 60k. But for general purpose it shows the point.
Through years we were making tricks to make viewer believe that he is looking at more complex models. And the industry get good at it - normal maps, screen space subsurface scattering, PBR, AO, proper mesh optimisation.
There are allegedly 100k polygons for Aloys hair alone, more polygons in Aloys hair than the entire character from past gen.
I believe it was more like 194k for hairs but it's LOD0 used in cut-scenes and maybe photo mode. Still you can see there's not enough polygons on hairs, especially closer to face (which could be proper optimised - more polygons closer to head and less at the back of the head).
Optimising by replacing 3D geometry with more efficient baked textures instead, like specular maps, bump or normal maps isn't a bad idea but we are still getting noticeable improvements with higher polygon budgets, I would argue that's where this gen has seen a big noticeable demonstration in Nanite. That improved geometry detail is noticeable
We are at the point where 100k polygons is quite sufficient number for character model (depends on character - his clothing, hairstyle, ect.). Sure if you'll be picky then you will find things to improve, but generally speaking you don't need much more in many situations. Normal maps served us well for long time and will be still because they are efficient. Then again there is a difference in scale of details that were baked to normal map back in Xbox era to what it is baked now. Back then you need to make a lot of geometry, now you are leaving it more for details (like texture of the materials), and less for bigger shapes. And that's good because real geometry and normal map caches light differently - normal maps will have more leaking lights because it's not real geometry, so it approximates the lighting information.
As for Nanite and more polygons - well it still doesn't work on skinned meshes and who knows if it will be or is it even needed. Working with hundreds of millions of polygons in programs like Zbrush is fine (well to the point because performance can be not good), but those models need to be skinned to rigs for animation. Working on dense wire frame in software like Maya is painful and at some point not even possible. Not to mention importing such big models to the programs or then animating it. Also those files are heavy - like model itself is like 20 GB (but probbaly can be compressed) in OBJ/FBX.
In offline rendering you are still using less complex models but then durring rendering they have dispalcement and tesallation applied. Maybe there could be better tools for managin this but for now they still aren't here and I don't believe they will be in forsible future.
when it comes to face deformations you always want a higher poly count
About 30k is enough and still we could work with less. More important are good blend shapes (geometry plus normal maps).
because you can only see that shit in photo mode against the sun zoomed in 300%
Can't agree on that. You can see it in normal cut-scenes, espceially that Guerrilla is using this perfect lighting setups for most of the game (so it emphesize the effect). Sure they are more visible when you zoom in photo mode but it's not like you see it only there. They were added for a reason and it's not only for photo mode.
As for the difference between generations, PS4 and XO had not only better poly counts or lighting, but also it introduced new rendering solutions like PBR. It was a real game changer and made asset/characters production more guided then previously. It provided nice jump in general fidelity (although without proper lighting setup it can be rather flat and underwhelming, because we don't bake into textures things like AO, cavity, lighting, etc.).
I do agree though, that PS4Pro and XOX made the jump to PS5 and XSX less noticeable. Without upgraded consoles we would get from 1080p to 2160p (targets) this gen. It wouldn't be a huge difference but still it would be something for marketing.
This generation has Ray Tracing yet it is still in it's infancy because of the power that it needs. SSD's were supposed to be this great feature but for now they aren't used properly by the engines and the gameplay design. The rest of graphical features is just an improvemet on what we already had.
.
). Then you need to make na UV for each strand, so it'll be a lot of them. On top of that you would like to optimise it so that the parts that aren't visible are removed (so less polygons and less UV space needed). This type of approach makes the process even more time consuming so it's not only making the polygon and texture budgets bigger but also time (and by that money).