Accurate Human Rendering in Game [2017]

Status
Not open for further replies.
One story related NPC from H:ZD

horizonzerodawn_20170o1xht.png
 
Main differences / advantages for the Logan head:
- All the surface detail is probably geometry, achieved with tessellation and displacement. This would be even more evident in close-ups, where a game character would have very smooth facial features and the high frequency details (skin pores, wrinkles etc) are recreated with normal maps only.
-Hair and fur are actual individual strands in geometry, instead of alpha mapped planes and such.
- Everything is raytraced: lighting, shadow, shading - so everything will look more realistic. One very obvious example is how the ingame eyes lack the refraction of the cornea and thus they look very different from a real eye:
1.jpg

The pupil is actually deeper inside behind the lens, but the strong refraction makes it look like it's closer to the surface.

So, basically, the CGI head is rendered with methods that closely simulate the real world physics, whereas the ingame renders are using more or less rough approximations, like screen space shading, cube map reflections, normal maps, shadow buffers etc. The difference between the computational requirements is several orders of magnitude.

This means that I don't really expect the next console generation's hardware to close the gap and make the move to doing things "right". There is indeed a lot of room for more advanced approximations that could get even closer to the CGI results, but I don't really expect to see real hair strands for the hair, fur, eyebrows, eyelashes; or proper raytraced refractions and SSS.

Witch zero
The demo renders over 63 million polygons per frame, uses "8K by 8K" resolution textures, and her hair is rendered with over 50 shaders, with each strand of hair rendered as a polygon.-wiki
By individual strand as a polygon I assume they mean individual polygonal strands.

There was also a paper showing realtime individual fiber for clothing might also be possible, but not sure the limitations. LOD should greatly help, as that would only occur during close ups.

 
Common thing about such tech papers is that it takes years to have them appear in actual games, at best - or they may completely disappear at the worst :(

23_nalu_02.jpg
 
nVidia just need to ninja their way into Team Ninja's dev house.

Wonder how it compares to TressFX tbh. Caveats, trade-offs etc.
 
Common thing about such tech papers is that it takes years to have them appear in actual games, at best - or they may completely disappear at the worst :(

23_nalu_02.jpg
Yup, but honestly, I don't understand why it takes so long for us to see such things in current videogames. Isn't that demo from ~2004?
EDIT: of course, I understand that this is a matter of resources assignment, but I think it's also time for the developers/graphics cards manufacturers to focus a bit more in these things that will make videogame characters look more real. I for one would prefer to see real-time hair/clothing instead of superultrahigh textures and superfancy shaders on characters. I mean texture quality, polygon density and shaders are quite good, by now, so please try to push other aspects rather than offering yet another SSS technique, which I will barely notice while playing, for instance.

I guess not many games are underwater mermaid simulators :(
That's unfortunate, you know? (I love mermaids)
 
Last edited:
You also have to consider that any tech demo will only demonstrate the tech in isolation and under complete control. Getting it to work in an interactive environment, and making it robust is a whole different challenge.

On the brighter side, there are some advancements happening, particularly with cloth sims - it is now a common expectation to have some level of this going on, at least for the player's character. Aloy for example has many pieces with secondary dynamics, and that also includes some of her hair (even if those are just poly-based braids). I wonder if Compute shaders can be utilized to increase complexity and/or leverage the tech to secondary characters; it is for example quite puzzling to see most of the NPCs in Horizon with parts of their costumes and hair intersecting everywhere.

Oh and I've just seen the close-ups from Aloy's skin - those are quite evidently based on scans too, and work reasonably well. But I still think that her face is unnaturally smooth :(
 
But I still think that her face is unnaturally smooth :(

I mean...

If you're going to genetically resucitate a hero who sacrificed her entire life (both her latest years and her life life) to save the world, with practically the only intention to ask her to sacrifice everything for the world yet again... you might as well give her flawless skin. I think it's fair. :D
 
Re: Horizon, I did notice that for a tribal society who live in mountains without any electricity to speak of (even though they have millions of robots running around running on electricity), almost everyone looked pretty immaculate and out of some Paris fashion week show.
That's fresh air and exercise!
 
Re: Horizon, I did notice that for a tribal society who live in mountains without any electricity to speak of (even though they have millions of robots running around running on electricity), almost everyone looked pretty immaculate and out of some Paris fashion week show.

Maybe they found old stocks of cosmetic creams and beauty products
 

Playing around with the time of day slider close to Aloys face (that's the same place as the screenshots I posted some days ago)
 
Finalised human skin shader for Squadron 42:
Human_skin_shader_head.jpg

"Heads were also successfully converted to use the human skin shader developed by the graphics team. Since there are 44 different areas of blended wrinkles and blended diffuse, the texture cost was quite high, at about 100 MB per head. With this change, roughly 90% of the original texture memory cost was saved without discernable visual impact."
 
Status
Not open for further replies.
Back
Top