Naughty Dog working on million poly game ready characters?

iWeReZ

http://playstationlifestyle.net/2012/01/20/naughty-dog-working-on-next-gen-game-with-million-poly-character-models/

So i was just wondering what you guys thought on this? is it even technically possible or a typo? i think it could read either, million poly game ready models, or taking those million poly models and making them game ready.

Would that mean the baking of normal maps to cease? or maybe higher resolution normal maps for the high poly models?

Ive seen what poeple here can do with 1,000 polys, so having that much to spend seems amazing,

Snefer

I think its quite obvious that they are talking about the highpolys for making normalmaps : )

Del

Snefer wrote: »

I think its quite obvious that they are talking about the highpolys for making normalmaps : )

this^

the quote they picked says: “this digital sculptor will be responsible for making million poly models game ready"

Everybody I know would assume that to mean: "bake a normal map to a game res model from your sculpt"

I think the article writer is just unfamiliar with dev terms.

Sandro

Their character models in uncharted seemed to be heavier than your average in-game character.

http://area.autodesk.com/userdata/image/inhouse/w/4a527-b3xij-2a32s-7u75t.jpg

Pretty dense geometry, almost pre-rendered cinematic kind of dense.

Million polies is probably over-exaggeration (and I can't imagine skinning and handling million polygons in any convenient manner) but since their titles seem to be so character-oriented I would not be surprised if they threw even more geometry to character models in their next game.

MainManiac

this digital sculptor will be responsible for making [his] million poly sculpts game compatible

iWeReZ

Ahh , thought as much But then, what will it take for a million polys to be the norm of a game character? considering the growth since PS1 era models.

Ged

Im guessing even naughty dog use lods of characters sometimes too. They cant be as detailed as that lady all the time! surely.

iWerez it will probably take 1 or 2 more generation of consoles or 5 years before its possible, doesnt mean its a good idea though.

Bibendum

Even if million poly models were the norm I doubt anyone would ever bother to model them that way, they'd more than likely model them with simple/clean topology that subdivides well and let the engine tessellate it with a displacement map.

ambershee

There are games out there on current-gen consoles that do actually include meshes with as many as a 1.5 million triangles. The caveat is that they are agressively handled with specialist level of detail systems.

In this case though, this is a simple case of the author jumping the gun, not doing any research and therefore not understanding his sources.

leslievdb

Dylan Brady

LOV U GUUS
But no somewhere said that the characters in UC 3 were ~40k tris

Stinger88

Do they even use +1million poly models in film. Say avatar. I think it would be overkill. Screen resolution only goes so far.

Unless every part of a model is going to get super close to the camera. Then its pointless.

oglu

let me dream...

sculpting a char with several million poly... bake some vectordisplacement maps into ptex...
mixed with realtime tesselation to get 1 million poly in closeup...

dream end...

amd is working on hardware support for realtime ptex and megatexture stuff...
http://www.anandtech.com/Gallery/Album/1601#1
http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/6

Ark

Well in Film, hero characters basically use the same setup as games except they use Displacement maps, sometimes in combination with Normal/Height maps.
Rigging a few million poly character is not something i think i would like to do. Thats why they use sub-division surfaces with the above maps.

Imho, engines are already pushing enough polies, especially with tessellation that character can already have good silhouettes.

Id love to see lighting and shaders pushed even further, over trying to render more polies on screen.

Proper GI and Physically correct illumination models make much more believe scenes than rounding off some characters finger.

Del

Stinger88 wrote: »

Do they even use +1million poly models in film. Say avatar. I think it would be overkill. Screen resolution only goes so far.

Unless every part of a model is going to get super close to the camera. Then its pointless.

they probably do go over 1mil
the high count is probably more to do with the quality of the render. The higher your polycount, the crisper the shader responds to it. Things like specular really respond to seemingly unnecessary polygon resolution.

oglu

an avatar char will have about 200 2k textures or even more... that will end up in a displacement texel ressolution of 800 million... say they use only 30% of the texture space... so you will have 240 million texel data to displace the geo during rendering... and thats only the handpainted stuff...

if you store the displacement data into textures you could do similar also ingame...
refering to my post above... texture streaming shouldnt be the problem annymore...

Razgriz

I think the funny part here is that the guy who wrote this article clearly has no idea what he's talking about, and is simply misreading Naughty Dog's job postings (while anyone actually qualified for the position would know exactly what million-poly sculpts are for), and all of the comments are normal gamers overwhelmed at artist-jabber they don't understand, wildly speculating as to the implications this has for next-next-next-gen games.

Ace-Angel

PS1 pushed around a couple of K's at most, pushing near the double digits in PS2 was true resource management muscle.

The in PS3 era, we get at most 30-40K for a character, this the MC in the grill type of setting.

So by the time the PS4 rolls out, we should be around the 70-80K limit for character.

Please not I'm talking about vertices with bone and other added stuff to make it feasible.

In UDK, you can already run into the 72K limit they put per mesh, so I'm not sure what the consoles can do to even catch up to this.

IchII3D

There was a studio insight video about the new game they are working on which talked about character polygon count and how extreme it is. But he also states that its a cinematic model, I'm assuming there cinematic characters will be crazy high poly, but when in game it will be LOD down. Almost like racing games, how the vehicle selection screen is often higher poly than when actually racing.

Snefer

Ace-Angel: We will go higher than 70-80k for nextgen, atleast occiasionally. I know of characters for this gen who had almost 70k triangles : ))

eld

Not to forget, hardware growth is almost always exponential, not linear.

We're not going to get double the amount of polygons for character, we'll get several amount of doubles, but then again, that is also irrelevant as we'll focus more on workable meshes using proper tessellation and shaders to add detail.

So I don't believe we'll suddenly start making these heavy and hard to work with meshes when we just as well can start pushing tessellation to bring in a non-polygonal silhouette and shape.

Also consider that there's no definite amount of polygons you use for a character, as it always varies depending on how many you'll have in your scene, which in this generation has been from 1 up to 1 mirrion.

Blaizer

And here we go... subdiv FTW.

r_fletch_r

how the hell do you skin and morph a 1 million tri character at 60fps?

Andreas

r_fletch_r wrote: »

how the hell do you skin and morph a 1 million tri character at 60fps?

Typically you only do it once, you'd have to be pretty fast to do it at 60fps

arf arf

Michael Knubben

Eld: Doubling IS exponential. Linear would be if the same amount was added each time.

Another thing to consider is that launch titles never use the hardware as efficiently as those that come out in the end of a console's lifecycle.
Because the current generation has lasted quite a few years, there will certainly be a big jump in possibilities, but nothing compared to what will be possible once companies have more experience with the platform.

eld

True, I just failed math

I guess what I wanted to say was that it'll be a bit more than 'just' twice as much polygons considering the console hardware was set over 5 years ago.

Snader

Nah. I reckon next-gen characters will be around 100.000 tris per piece, but all shaders will be more complex and the environment art & scale will get a huge boost. For example, look at the draw distance differences between CryEngine 2 (PC) and 3 (consoles) in this video:

[ame="http://www.youtube.com/watch?v=kRWRskYLdno&hd=1"]CryEngine 3 Dissection and Comparion HD - YouTube[/ame]


Also, a very recent game, RAGE, seems to still use just a fullbright diffuse for the skybox. No movement, nothing. Just a huge wallpaper. =( We seriously need some more detail in our console worlds.

ScudzAlmighty

Wasn't the Rage stuff a result of the MegaTexture implementaion though? Where the lighting on everything was baked in to the texture and also why some objects (like the infamous microscope) look so bad up close? I'd be really curious to hear from some of the iD'ers if it was really all that helpful/worth it over traditional texturing methods. I'm playing it now and while the atmosphere is excellent and the environment looks amazing as you drive by, I really feel like nothing holds up well to the kind of close-up inspection an fps naturally presents.

Also, the comments on the op article are just so... I guess it's the authors fault not understanding what he was reporting but... damn some people are stupid.

rv_el

Man those ND guys are so crazy with pushing polys. They have had amazing numbers this generation.

One cool thing to know about Uncharted is they do not LOD anything really. They say its more of a cost to do LOD's than it is to just leave it well enough alone. Pretty crazy. Dunno if that changed in UC3 but I think it was that way for 2.

Bigjohn

From what I understand, a major problem with tri count on characters is having that many skinned vertices. For every vert you add, that's more than just increasing the vert-count, it has to have skinning information too.

So I wonder if we'll actually see it double.

Razgriz

Bigjohn wrote: »

From what I understand, a major problem with tri count on characters is having that many skinned vertices. For every vert you add, that's more than just increasing the vert-count, it has to have skinning information too.

That's a good point, but it really only applies to characters and deforming meshes - and those have already gotten pretty high, if you look at the Uncharted or Gears LOD0's. I could see a lot of possibility for insane mechanical/environmental scenes that are very complex, with lots of stuff that breaks this silhouette. It seems like that's getting easier and easier to do without the sheer number of tris causing any sort of a bottleneck.

Autocon

Drake is around 50k tris if I recall correctly. But we dont in tris, but our engine counts things in UV's which get broken up by smoothing groups, shader assignments, uv islands and so forth.

And for the most part we dont use LOD's that is true. There are some things that have them, generally its things like vehicles and such. For environments we used progressive meshes to a very slight extent. Really depended on the level, the artist and the asset they had if they wanted to use them/they would work for there level. Things like fade out were more relied upon then LOD's.

For example, I never used them at all for my level. Having LOD's doubled the memory cost per asset for the overall level memory and every single one of my levels was at the cap and took months to even get it within runable rates (silly big levels haha).