"Yes, but how many polygons?" An artist blog entry with interesting numbers

So Chloe ended up having more polygon than Drake himself in max case scenario, well being a female character means this is obvious. And it seems there is no LOD going on for Drake in SP.
 
So Drake in highest LOD 37k polys in U2... seems I wasn't crazy then with multipass talk. Well less atleast. :LOL:

That still does not confirm anything about multipass... But 80K did sound too much, I still stand by my first guess that it was meant to be 18K on average (ie. not the main cast).
 
Some Uncharted 2 lovin'

http://www.zbrushcentral.com/showthread.php?t=79105&page=1&pp=15

Edit: and some relatively exact poly counts from Rich Diamant, lead character artist:

the wigs ranged a bit dependent on the character.. Drake's hair is around 4k tri's.. chloe's is around the same..

our lod's were originally hand modeled.. it was annoying.. we then went with an automatic PM solution which is what you see in game.. Drake however never gets LOD.. except for multiplayer.. but single player is the high resolution model the whole time.. same model that is in the cinematics.

the models ranged a bit as well.. drake was around 37k .. chloe was at 45k at her highest... each level of LOD was cut in half with 3 different levels. lowest being around 2500 poly's..

if any of you guys will be at GDC i'm doing a talk which will cover a lot of that..
sometimes I would texture the high poly before going onto the low poly.. however, this time around I tried a slightly different approach of sampling first, then texturing the low poly.. it works pretty well.. however, it makes it harder when big changes happen or when you need to show off the work.. I think next time around I will stick with doing everything on the high and transferring to the low..



I hope ND remains true to their selves and post that GDC presentation on their website eventually ;)
Interesting that they texture the high res first - Ninja Theory did the same with Heavenly Sword, and while I see some reason in this, it also means a lot of extra work and ignoring some more efficient techniques.

Damn Laa-Yosh you beat me to it, yes nathan is around 37k.

i had just now found the link.

that was a good find, in a short a mount of time too, good job.:smile2:

Edit, and also i like to say that let us not forget that these high LOD models are most likely used for the cut-scenes, as some of the in-game shots i displayed showed a difference in level of detail.
 
Last edited by a moderator:
It is the dev that created the models that said it himself that only Drake is samein cutscene (U2) and ingame. Doesn't get more clear than that. :smile:

Where does it say that? I guess it's too clear for me. I read that Drake's character wasn't LOD'ed, nothing about all other characters having different models for cutscenes. [/shrug]
 
This part. Actually it might be highres model for all ingame! XD
our lod's were originally hand modeled.. it was annoying.. we then went with an automatic PM solution which is what you see in game.. Drake however never gets LOD.. except for multiplayer.. but single player is the high resolution model the whole time.. same model that is in the cinematics..
 
Could be that they do one LOD automatically for in game characters other than Drake, it just doesn't say that.

Btw, I wonder what Chloe would say about it taking an extra 8000 triangles to cover that ass. :LOL:
 
37K sounds real good. I can imagine the fun the Lead artist would have when he builds a base Drake model say around 9.5k then just hit Meshsmooth and vola:) then tweaks from there.
 
This part. Actually it might be highres model for all ingame! XD

No wonder Drake's model is so damn detailed in SP.
angry.gif
 
Don't know if this has been posted before but here it goes anyway.

Halo: Reach is being propagated as the next game to use technology that packs a wallop and this is what is being discussed as being put into the new engine that is being developed for the game. These features were spoken extensively at this years GDC conference with reference to DX11 technology:

-Tessellators with close to 30K characters to achieve cinematic quality features.
-The Hull Shader to patch control points in order to modify the behavioral pattern of the tessellator.
-Domain Shader to optimize performance of the output of the GPU in order to mitigate bandwidth use.

These features allow real-time display of real-time images which are otherwise not achieved making creation of objects in the world easier for programmers whilst saving up on a limited memory module. Silhouette elaboration and terrain rendering whose objects offer a healthy degree of curvature.
All this allows for a visually heavy game in the Halo series. More as time elapses.

http://www.gameguru.in/first-person-shooters/2009/11/the-technology-powering-halo-reach/
 
Characters don't look like they're using tesselation, all that hard surface detail in the armor has to be pretty tightly controlled to work well with normal mapping. Faces maybe, but it's exactly the place where you want more actual vertices to have better control over bone weighting for animation.

I'd say they're using tesselation on the terrain. It did look kinda smooth and nice.
 
Maybe cutscene models for main characters? They do look much better than in the other Halo games (3 and ODST) I'd say pretty comparable to Mass Effect characters wich are around 20k right?
 
Last edited by a moderator:
Most of the armor pieces are a collection of flat planes with various transitions that have to be very tightly controlled for the normal mapping process to work well. Building hard surface stuff for games is a lot more complicated then pure organics (humans, monsters, clothing, etc). Tesselating these transitions between the planes would probably mess up the tangent space that the normal map tries to work with, resulting in shading artifacts. The flat surfaces would look OK, but why would you want to add more polygons there?

And again, adding more geometry to the faces would only work if they already have enough vertices for weighting the bones they use for facial animation. Take a look at the Uncharted 2 character wireframes to see how far it might have to be taken to work well. Tesselating those meshes wouldn't add any visible difference even in the cinematics.

On the other hand, it makes a lot of sense to tesselate the terrain with an adaptive, view-dependent scheme. That could add more polygons to the parts that are closer to the camera (player), increasing the detail and variation of the scenery. We haven't seen enough yet to decide but I believe this to be the case, based on simple logic.
 
Back
Top