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

[assen]

But if you're still spending an age on modelling, the added expense of real life models strikes me as wasteful.

It's probably easier to leaf through a thick portfolio of model photos to find interesting faces than to paint them from imagination; it's a common thing to do.

Re: 2048x4096 vs 2x2048x2048: it's easier to manage square, same-sized textures, than to special-case for rectangular, and for something like characters in an FPS, the added drawcall isn't going to be a problem.

 

[semitope]

Is this a sensible, economical approach? To my eyes, KZ2's characters aren't any better looking or more realistic than other titles, so what's the point in hiring models? If you could scan them and create a model in minutes/hours, it'd be worthwhile. But if you're still spending an age on modelling, the added expense of real life models strikes me as wasteful.

It is interesting though. I look at it as being similar to movie casting and I believe from the U2 example there is something to be said for basing your character designs off actual organic beings. At the very least it saves on the imagination process and finding a thought up face that fits your requirements

 

[Shifty Geezer]

It's probably easier to leaf through a thick portfolio of model photos to find interesting faces than to paint them from imagination; it's a common thing to do.

Why pay models when you can grab inspiration off the internet for free?

 

[Laa-Yosh]

The variation in realistic and characteristic human faces is very difficult to reproduce. We almost always use some kind of photo reference as well, although not in such a strict way that Guerilla and MB did. They've made several high res photographs, from all angles, and followed them as precisely as possible in the modeling. They've also projected the images onto the model as well, as a basis for the color textures.
We on the other hand just look at photographs for inspiration and leave our artists to experiment. I'd say a complete hero human head takes 7-9 days on average for us, but we split up tasks between differently specialized artists.

You're also right in that the actual ingame models are not as good looking as one would expect from such schedules and effort. But based on the images of the source artwork in this book I can definitely say that those are very, very high quality, with a lot more unique detail then what Gears had (which builds heavily on repetition). The trouble seems to be that they had to reduce polygon counts and particularly texture resolution a bit too much.

As for outsourcing, think about it like this. Massive Black in China maintains a staff of a few dozen character artists (and probably a lot more) that they can use for a lot of different projects and fill up their schedules completely. Hiring 20 additional character artists at Guerilla would get problematic as soon as they complete the content for the game, as they'd then run out of work completely.
It's also quite likely that salaries, benefits, office space and such are far more expensive for an artist in the Netherlands compared to China.

I'm still a bit surprised about these figures though, considering that Epic can put together a Gears game with about 5-10 character artists at most. It's probably that all the Helghast have too similar looks and you don't notice variations...?


Anyway, what I wanted to say is that outsourcing can definitely make economic sense, it's the actual time for KZ2 models that's a bit surprising indeed.

 

[Laa-Yosh]

It is interesting though. I look at it as being similar to movie casting and I believe from the U2 example there is something to be said for basing your character designs off actual organic beings.

Uncharted 2 characters have absolutely nothing in common with the voice/mocap actors. People sometimes mention Elena but even she's completely different from the actress, not to mention Claudia Black / Chloe or Nathan and the guy (can't remember his name).

 

[Laa-Yosh]

It's probably easier to leaf through a thick portfolio of model photos to find interesting faces than to paint them from imagination; it's a common thing to do.

Yeah, we had such an issue...

"And we want him to look like *** *** (hollywood actor)."

"Uh-oh, he looks too much like *** ***, we gonna get sued, make him look less like the guy."

Would've been a lot easier with a model we'd be allowed to use 1:1

 

[Laa-Yosh]

Why pay models when you can grab inspiration off the internet for free?

Might take too much time to search... It's also that GG used an exact likeness, complete with photo texture on the face, which can turn into a legal issue. Also, this way they had exact reference to send to the outsourcing artists, a lot easier and streamlines communication and reviewing.
All the rest of the character designs were highly detailed drawings with multiple vies, too. They did not leave much for the imagination of the artists, which can actually be a good thing.

Not that everyone does what GG did, but it is a valid approach.

 

[semitope]

Uncharted 2 characters have absolutely nothing in common with the voice/mocap actors. People sometimes mention Elena but even she's completely different from the actress, not to mention Claudia Black / Chloe or Nathan and the guy (can't remember his name).

Thats a huge statement to make. The fact that their hair colours are the same alone, however minor it is, completely disproves that particular statement. There are similarities but it would become a chicken and egg situation. "which came first?" Did they find a person to fit their designed character or did they find an actor they thought fit the role and ended up basing the character (facially in some way but mainly expressions and movement) on that actor.

I agree with your other 3 posts

Could this discussion on modeling characters after real life individuals be spawned in another thread? It would be interesting to compare differences in faces and such.

 

[ultragpu]

This is a lot more complicated then that.

Originally the Sniper's been UVd so that it sits on one big quadrangular map, however the UVs are also divided into 4 equal sized quadrants so that the 1 texture map can be split up into 4 smaller textures if necessary. In the end though, it used two layouts, one for the head and one for the rest.

Now, there are no limits on how many maps you can use. However as far as I know, a single draw call can use only one set of textures and one shader, so if you split your model up, you need to draw it in more calls. Which is why it's rare to use more then two.
I'm not sure why most studios prefer for example two 2048*2048 maps instead of one 2048*4096 though.

No word about texture resolution in the game, but that's OK. 1-2 x 2K is the most common texture size for such characters anyway.

Yeah, seems like human characters use a wrinkle map on top of blendshapes for facial animation. They've written a custom Maya node for it (want), but only use one layer (Uncharted has two, I might need three). They're using Maya's shader editor for all of their custom stuff as it seems.

.

Ah I see, thanks for the explanation. Also the wrinkle map is doing quite a justice to the believability of the model, I wonder if they simply made a greyscle map in Photoshop and converted it to a Normalmap or just projected it in Zbrush only with wrinkles?
On the subject of outsourcing to Massive Black Shanghai studio I believe it's mostly cost reduction to a huge margin speaking from someone who was born there. You just can't believe how much labour fee you can save compared to other places.

 

[ultragpu]

My dear Mods, just wonder if this is appropriate to post here.
OK guys, I can't help it by would like you all to have a crack at this WOW character Tauren model I made. I always wanted to picture how a WOW character would look with current gen spec, the design is based on a toy named Korg High mountain from WOW with slight modeification. So here it is.
http://i50.tinypic.com/x22zqr.jpg
http://i48.tinypic.com/2hs9huf.jpg
So it's got 10k polys, two 2k * 2k texture maps. Zbrush for high res model, Normalmap and cavitymap. I'll post wireframe and maps if interested, cheers .

 

[COPS N RAPPERS]

hmmmmm, i can pretty much agree with shifty geezer, and plus to boot, there wouldn't be much of a benefit of using real models if the game that you're building it for is just going scale the model down heavily.

there's a program out there that generates faces at the drop of a hat, what takes these people to do in months can take them just an hr or two. http://www.facegen.com/
export your stuff to either maya or zbush and you can create some crafty work.

it's true finding people would be the faster way, (if you want to base your character on real people and have a person in mind.) but i think nothing beats a little self inspiration, as an artist i rather take full credit for my work.....if it'd meant for a bigger paycheck.:smile2:

 

[Laa-Yosh]

I still don't believe in procedural human character generators. Facegen has been used on Halo ODST and pretty much everyone agrees that the main characters did not really look good. The quality you get from real living artists is always going to be significantly better.

 

[Shifty Geezer]

As a different perspective, I don't see much difference myself. The models are neither here nor there. If you use Facegen to create a basic model, and then let an artist tweak it to add character, coupled with an effective renderer it'll be good enough I think, certainly for the majority of gamers. The cost/gains ratio favours shortcuts IMO.

 

[COPS N RAPPERS]

As a different perspective, I don't see much difference myself. The models are neither here nor there. If you use Facegen to create a basic model, and then let an artist tweak it to add character, coupled with an effective renderer it'll be good enough I think, certainly for the majority of gamers. The cost/gains ratio favours shortcuts IMO.

yes, the results would be similar and pretty good. if you're handling Facegen as a stand alone program then your results would actually be pretty generic and a bit lifeless. the program was meant to work in conjunction with other programs.

i do see where Laa-Yosh is going though. he's viewing it on a bigger scale, where you're working with a studio that has all the real expensive equipment at hand, such as a mocap suite that grabs all facial coordinates as well as animations and whatnot. if you have all that then no you wouldn't see much of the benefit in using a generator.

the real importance of the program is also for small budget companies, the companies that roughly have few people working in character modeling and designing, using the program would save them great costs than rushing over to a real studio and borrowing their staff.

never the less, if you have an eye for detail you can make the program work in your favor.

 

[Neb]

I've seen stunning faces done for games (mods) with facegen. Though the best ones used real face image(s) as base.

 

[joker454]

400k...combined from all passes or purely mesh alone?

400k mesh is 10x what a current gen car game is using for a base mesh. and more than twice what GT P uses.

source?

I came across this video of someones render reel:

http://www.youtube.com/watch?v=AOMVCpiKAHI&feature=player_embedded

It's a cool vid with nice looking cars, and he gives poly counts of all the cars at the 5:11 mark. Quick numbers summary, the Mustang is 71k, Camaro is 33k, Jeep is 84k, and the Vaz is 33k. It's not really an answer to GT poly queries, but I figure that video can give people a good idea as to how many poly's make a nice looking car.

 

[Laa-Yosh]

It's a cool vid with nice looking cars, and he gives poly counts of all the cars at the 5:11 mark. Quick numbers summary, the Mustang is 71k, Camaro is 33k, Jeep is 84k, and the Vaz is 33k. It's not really an answer to GT poly queries, but I figure that video can give people a good idea as to how many poly's make a nice looking car.

Those cars use subdivs, they're tesselated for the actual rendering. So the real triangle counts are probably 8 or 32 times as much, depending on how close they are to the camera.

 

[joker454]

Those cars use subdivs, they're tesselated for the actual rendering. So the real triangle counts are probably 8 or 32 times as much, depending on how close they are to the camera.

Ah ok, my bad. I guess then re-purpose that vid then as an argument for mandatory tesselation hardware for next gen

 

[proffesso]

www.polygonpirates.com

those are realtime cars, in the 25 - 40k (exterior) range

 

[gabe2500]

how does zbrush 3.5 come into the equation? Since now you can export hd geometry for maps, would that make in-game models look that much better?