Accurate human rendering in game [2014-2016]

Status
Not open for further replies.
This is great stuff. And according to the guys doing the presentation, all realtime.
 

Attachments

  • Screen Shot 2014-12-09 at 23.35.08.png
    Screen Shot 2014-12-09 at 23.35.08.png
    2 MB · Views: 29
  • Screen Shot 2014-12-09 at 23.37.22.png
    Screen Shot 2014-12-09 at 23.37.22.png
    1.9 MB · Views: 31
  • Screen Shot 2014-12-09 at 23.45.24.png
    Screen Shot 2014-12-09 at 23.45.24.png
    1.4 MB · Views: 33
  • Screen Shot 2014-12-09 at 23.45.40.png
    Screen Shot 2014-12-09 at 23.45.40.png
    1.2 MB · Views: 33
Last edited:
Most of the stuff they're doing is basically a 1:1 of what we do, except we don't have to cater to GPU and console hardware limitations. Like, they're using wrinkle maps for the facial expressions, whereas we can build more dense face meshes that can accommodate the deformations without any help from textures.

Also interesting is that they're using several extra texture layers to add cloth fabric patterns and stitches; makes sense with realtime limitations, we can always use a bunch of 4K maps instead and paint everything by hand, whereas they still need to conserve memory.

Also, faces seem to have around 10-20k quad polygons, very close to us - we have 30-50k polygons, but we don't re-use the models for hero characters, each face has it's own polygon layout.

I guess this is as close as it can get to offline CGI and movie stuff. Pretty damn amazing!
 
They also specified why they don't scan characters' faces - or at least Nathan. The amount of detail that is basically hand modelled and drawn is beyond my comprehension, to me he basically looks real.
 
Last edited:
After watching that presentation with the shirt dynamics, hair movement, etc (stuff I didn't notice the first few times I saw the gameplay demo) reminded me of Jason Rubin's (co founder of ND, in case you don't know who that is) "Great Game Graphics... Who Cares?" presentation at GDC in 2003. The contrast is pretty funny when Jason talks about, "so we model each individual hair and it moves like in real life... so what, does that make our game better to play?" and then to watch the Drake video with the panel taking about the hair, etc.

Good thing the star of the gameplay demo (IMO) was the new additions to the gameplay rather than the graphics (which are quite good btw).

For those interested:
http://www.gdcvault.com/play/1016690/Great-Game-Graphics-Who-Cares

The talk is mainly about the resources needed to get marginal gains in graphic fidelity till you reach a point (economically) where graphics cease to wow people enough to sell the software on its own and you need all parts of the game to be well done to stand out.
 
Also, faces seem to have around 10-20k quad polygons, very close to us - we have 30-50k polygons, but we don't re-use the models for hero characters, each face has it's own polygon layout.

I guess this is as close as it can get to offline CGI and movie stuff. Pretty damn amazing!
Veyr nice insight Laa Yosh, also does 20k quad poly mean 40k tris by the way? if so it's an incredible amount for the face alone!
 
After watching that presentation with the shirt dynamics, hair movement, etc (stuff I didn't notice the first few times I saw the gameplay demo) reminded me of Jason Rubin's (co founder of ND, in case you don't know who that is) "Great Game Graphics... Who Cares?" presentation at GDC in 2003. The contrast is pretty funny when Jason talks about, "so we model each individual hair and it moves like in real life... so what, does that make our game better to play?" and then to watch the Drake video with the panel taking about the hair, etc.

Yeah, but Jason Rubin is a "lets make a game" guy, a these ~6 people are "we are so happy that naughty dog is so character-focused game studio" modelers.

Even if during 2003 Naughty Dog had GDC panel about game characters, those same modelers would still be excited about extremely small details that they were able to insert into the engine. Their job is modeling stuff and executing orders from the top, not thinking much besides their area of expertise.


Btw, I really liked that segment when they shown how face can move, opening mouth causes movements from the ears all the way down to the base of the neck. Great rigging and attention to details.
 
They also specified why they don't scan characters' faces - or at least Nathan.

Yeah, the general design approach to hero characters in games is to create a blend of known actors and celebrities whos' public image fits the desired result. It is theoretically possible to find a real living person who looks like that but obviously it's not practical, and it's far better to get a talented (digital) sculptor instead.

But I would like to point out that it's still quite valid to use a near 1:1 likeness as well. Hiring a known celebrity is still a good choice, especially an actor who can also lend the voice and the performance. Just look at the buzz that Kevin Spacey's involvement generated for AW - he has now made appearing in a video game cool and a 'real' job, so expect a lot more hollywood talent to appear in the future.
It's also working with people other than actors, as a lot of the actual character will be defined by the voice and the performance and it's also cheaper. A very good example would be Shepard in Mass Effect, where Mark Van der Loo's face has pretty much become a video game icon, but it also took Mark Meer's voice acting to really sell him. Funnily enough Bioware chose to go with a hand sculpted and not so realistic face with FemShep, I believe Jennifer Hale's performance should have deserved a more memorable face. If BW decides to release a 1080p remake of the trilogy for nextgen consoles, they should really look into casting someone for the 'official' face. And let's not forget that a large percentage of even ME1's cast was also based on scans, not to mention Yvonne Strachovsky for Miranda, some very good choices IMHO.

However in the case of UC4, I still think that not using scans had more to do with the general art style, where a completely realistic face would not work with the stylized rendering; and also with continuity, where such a sudden change in Drake's looks would've left gamers a bit disconnected from the character.

The amount of detail that is basically hand modelled and drawn is beyond my comprehension, to me he basically looks real.

I still think that UC4 Drake is too 'perfect' - far too much symmetry, and a lack of the subtle imperfections in the general forms. If you would meet a guy like that in real life, you'd be creeping out ;)
It's an interesting issue though, quite two-sided. We subconsciously prefer people with more symmetrical faces, as potential partners or friends/bosses - but 100% symmetry is impossible in nature and so it's as appalling as a highly asymmetrical face. Even the best looking celebrities have a lot of asymmetry, but it's almost impossible to see in photographs or real life, it's only when you look at an accurate CG model where it becomes visible. So a completely artificial character has to be very carefully balanced.
 
By the way, it'd be just so cool if you could meet Shepard again in a future ME game, that takes place about 10-15 years after the trilogy. Van der Loo has aged considerably since he's been scanned, but in a graceful way, he'd make a really charismatic and rugged older Shepard. And with today's advances, they could do a full set of scans for all the expressions, it could look amazing and work very well...

Mark-Vanderloo-Verino-FashionWeek-Madrid_TINIMA20120831_0350_5.jpg
 
Last edited:
After watching that presentation with the shirt dynamics, hair movement, etc (stuff I didn't notice the first few times I saw the gameplay demo) reminded me of Jason Rubin's (co founder of ND, in case you don't know who that is) "Great Game Graphics... Who Cares?" presentation at GDC in 2003.

Yeah, many of us were referring to that talk when talking about diminishing returns with the nextgen consoles some months ago.

The one thing he's not entirely right about though is that even these small visual steps seem to add a lot to video game characters; UC4, The Order, Quantum Break and the others are all able to push the envelope. But I'd argue that it's more about the acting abilities and a bit less about the looks, and also about the content creation advances.
 
I still think that UC4 Drake is too 'perfect' - far too much symmetry, and a lack of the subtle imperfections in the general forms. If you would meet a guy like that in real life, you'd be creeping out ;)
Out of all the things I would do if I met a guy that looked like him, creeping out would really not be one of them.
 
Heh, I'm sure you feel like that, it's a really good job of the 'ruggedly handsome' look indeed :)

But trust me on this one, a life sized version would be super creepy. Think Madame Tussauds and such, although the scariest thing would be if he could also move. I don't like mentioning the Uncanned Valley theory with regards to CG characters, but this is exactly the case - a lifelike robot of some sort that just isn't 100% real.
 
and also about the content creation advances
I've realized I know next to nothing about the content creation pipeline, do you know of any (free)resources online where I can read about how it currently works and where the art side of things are headed? Also I've heard the term technical artist thrown around a few times what's there role in things?
 
Well the ND presentation is a relatively good representation of what is done an how.

Game assets usually start with zbrush sculpts, then they model the ingame lowpoly asset on top of that and bake normal and other maps, then they paint the rest by hand. Shading is done in custom tools, rigging is in some off the shelf app like Maya or 3ds max. Exporters are always custom made, to handle additional metadata.

Technical artists are the guys familiar with the process of baking the texture maps and preparing the asset for the engine export by adding custom data and formatting. There are many intricacies there and they're usually engine specific, and traditional art guys tend to have trouble learning and/or understanding these things. It's also something that's more common with environments and props like weapons, for example think about how you can customize your guns in an FPS game with various textures, add-ons and other features, all of these require complex systems and carefully prepared assets. Or there are the randomized levels in Diablo where the various pieces need to fit together seamlessly, or instanced assets for foliage and so on. Or an open world game might have a complex system to build crowd characters from a base set of assets and implement color variations and patterns and such, once again a mixture of technical and artistic issues.

I'm not really into this stuff at a hands-on level though as we're not using any game engines and our projects require different solutions for different problems. Still, the best online resource has to be the Polycount.com forums, where every major game developer has artists posting stuff, lots of sticky threads full of info and links and so on.
 
As much as I like the new and improved Nathan Drake, I do think the animation does stick out now more than ever. And not in a good way I might add. There's just no weight to his movements, his legs flap about with a mind of their own when he climbs, he still suffers from over-emphasized, cartoony eyebrow action, and the way he handles the rock pick in particular just looks odd. Compared to how Lara Croft or Arno Dorian look when they climb about stuff, the way Nate gets around just looks outlandish and weird.
 
By the way, it'd be just so cool if you could meet Shepard again in a future ME game, that takes place about 10-15 years after the trilogy. Van der Loo has aged considerably since he's been scanned, but in a graceful way, he'd make a really charismatic and rugged older Shepard. And with today's advances, they could do a full set of scans for all the expressions, it could look amazing and work very well...

Mark-Vanderloo-Verino-FashionWeek-Madrid_TINIMA20120831_0350_5.jpg

Wot. Shepard kinda looks like Illusive Man now with that hair cut. (It's a twist :eek:)
 
As much as I like the new and improved Nathan Drake, I do think the animation does stick out now more than ever. And not in a good way I might add. There's just no weight to his movements, his legs flap about with a mind of their own when he climbs, he still suffers from over-emphasized, cartoony eyebrow action, and the way he handles the rock pick in particular just looks odd. Compared to how Lara Croft or Arno Dorian look when they climb about stuff, the way Nate gets around just looks outlandish and weird.

I guess that they are using previous UC animations; Sucker Punch re-used most of inFAMOUS 1&2 animations for Second Son and I believe that Ubisoft as well used some of the "old" AC animations" in Unity.

Now personally would I would like to see a "slower"/heavier Drake firstly because it would be more believable since he is older and secondly because now ND can deliver a more realistic/complete simulation but no doubt if they alter the animations climbing, running and shooting would not feel/be as in previous UC and I doubt that ND wants it.
 
Last edited:
It's not the speed really. Lara is just as fast. Heck, she's probably faster. And Arno basically scales buildings with the speed of a monkey. I think all the simulations and the animation blending is probably doing ND a disservice here. Lara doesn't have the same amount of animations. The ones she does have are way more convincing, though. TR Underworld had similar issues. Tons of animations, yet nowhere near as convincing in motion as TR Legends and TR Anniversary.
 
I can see what you're saying, Drake has always been more 'goofy' than other serious characters, but I think that's what makes him Drake. He's still more cartoony than other main characters even though ND strive for a realistic look. I love it, personally.
 
Status
Not open for further replies.
Back
Top