View Full Version : Polycounts: Whats the standard

12 December 2003, 08:56 PM
i know that polycounts vary depending on the type of game your making i.e. 1st person, 3rd person, stratergy etc but as i've been browsing through the different posts on this site i've noticed that the polycounts are really high (into the 6 & 7,000's) and was just wondering has this become an acceptable level for modelling at because i've been trying to stick to around 3,000 or less. I read that the main characters in games such as halflife 2 are reaching the 10,000 point (for main characters)

i know technology has been advancing but i didn't think they could be pushed this far so soon (especially given that the mainstream audiences don't live on the cutting edge of technology like the hardcore gamers etc)

just wanted to get other ppl's views on this subject =)

12 December 2003, 10:00 PM
I would like to know too! I've seen lots of debates on the number of polys used in Half life 2.

Have u seen Edge Magazine in the UK? In the preview of Transformers Armarda, its says Atari are using 12,000 polygon in-game models, and thats for the PS2.

12 December 2003, 10:04 PM
Just learn how to turn your high poly work into medium poly stuff. That's what everyone is doing now.

Hell, they did it in Quake 2. The original Quake 2 models were like 20,000 polys, and they crunched them wayyy down for the actual game.

Just about every first person shooter that is coming out next year will be using normalmaps, so get used to doing that.

If you're wanting to make a high quality character, then work as high as you want, and then bring it down and bake the high poly as a normal map. :)

For today's technology, I'd say a per character polycount of 6000 would be good enough to make a well detailed character, with a moving mouth and everything. It just depends on what you're shooting for. If your game is going to be a whacky shootemup, then it doesn't really matter. If it's going to be a moody, make you pee your pants type adventure, then it's best to add as much realism as possible.

12 December 2003, 10:16 PM
Sorry for my ignorance, but what are Normal Maps?

12 December 2003, 10:32 PM
yeah i did read that article in edge (get it every month because its the only games magazine worth the paper its printed on) but company's are renowned for boasting such things but we'll see.

normal mapping as far as i'm aware is rendering a high poly model and then using it as a kind of texture on a low poly model and im guessing because of the name that it does this by working on the normals visable to the camera however i only have a vague idea of what this is so i could be way off the mark here.

12 December 2003, 10:53 PM
basically taking detail from high rez models and making it display on a low poly model, so that it looks like the high rez. that's if you actually have a high rez mode. if you dont have a high rez character, or want to use it for level geometry, you have to paint them yourself.

hand painted normal maps are similar to software rendered bump maps, but can be used in realtime to effect how the light hits a surface. It makes a pixel seem closer to or farther from the actual geometry based on the color change of the neighboring pixels. this means that you can get all kinds of surface detail with a simple flat plane. normal maps are painted in the same way you would do a bump map for high-rez software renders, painted in grayscale. You then need to find a plug in for photoshop that allows you to use a filter that will conver the grayscale to red/blue/green, but i cant remember where i got mine. anyone have a link? also, if anyone has corrections to this, let me know as this is my understanding of it.

12 December 2003, 12:03 AM
Take a look at DOOM 3, if you want some examples of normalmaps in use. They're on everything from the walls, to the player models.

12 December 2003, 12:54 AM
dont want to sound like a complete fool. but what exactly are some of the better ways of getting you model down to mid/low polys? i mean do you just have to do it the long hard way or are there some tricks to speed up the process?

its just cause if your models really high poly then i would imagine it to be quite hard to do it all by hand...but i may be wrong.

12 December 2003, 01:14 AM
but the chars in doom 3 are up to 2000 the main chars are of course higher. When using normalmaps you will automatically have to come down witht he poylcount. I want to see how high Deus Ex 2 and Halfe Life 2 really go.

12 December 2003, 08:01 PM
As you can see I am new to these forums, so hi all. Anyway, HERE ( is a parfect example of this process at work. This was done with a 3Ds Max plug-in, called POLYBUMP. This is what ID used for DOOM3 and will soon be used in Farcry by Ubisoft. As you can see it does its job well, reducing a 250,000 poly model to 1500 polys and retains the detail very well. You can scroll through the screen shots to see more.

Also, they have a demo of the plug-in and manuals available HERE ( for Maya 4.0 and 4.5 and Max 4.0

Hope this helps a little ;)

12 December 2003, 02:58 PM
Polybump is not a mesh-optimizer. The two models are created by hand, then Polybump (like the other tools around) converts the surface data from the high-res model onto the low-res model.

There is no easy route for creating the low-res model. Mesh-optimizers do not do a good job, none of them maintain good topology.

Here's a thread with lots of normal mapping info, tools, tutorials.
Normal Mapping in 3ds max (

12 December 2003, 03:26 PM
Indeed some development houses make their models around 10K polys. Stalker is a good example. Some, however, prefer making their models around 5 - 7K. Some use normal maps, and some don't... I guess the avg amount is between 4K - 9K today. Just depends on how many characters the developers plans to have in view at one point in time. This also depends a lot on the texture resolution, because it actually takes up more space on the video card's memory. Normal maps are becoming more and more a standard though, so I suggest you learn how to work with them, if you plan to work on characters in future. If you want to check out some movies on this... look on Half Life 2 fan sites that have links with movies--most of them have a link to normal map modeling by VALVe.

Something else that greatly boosts the quality of the models in games, is something that is called TruForm. This method was developed by ATI to increase the amount of polygons on a model, without a major performance hit. You can make a 400 poly model look like 4000 without having to do anything at all. The developers of Raven Shield added that to the Unreal Engine. CroTeam did the same with Serious Sam: Second Encounter. BTW, their Serious Sam 2 Engine is going to be insane!

The bad thing about TruFrom, is that only ATI cards support that feature (I think.) If I am wrong please tell me!

There are tons of ways to make your characters and props in general look really good and realistic now. However it all comes down to the basics anyway. You probably need at least around 3K - 5K poly model, and an experienced texturer. If you have those two, you can throw on the normal maps after making a high resolution model to make it more realistic. At this point it should look pretty convincing (it won't if your environment doesn't look good.) Some games have excellent characters, but it's hard to focus on that, when the environment looks like it fell behind on quality a few years. I think it works the same way for the environments.

The best characters and environment combination that I've ever seen in a game is Knights of the Old Republic.

Jonny Bubonic
12 December 2003, 04:19 PM
If you use Maya, you might find this useful for reducing a model . It lets you assign "weights" to areas of importance, so that when you reduce the model down you can keep specific areas of detail that you don't want to lose. It's far better than Maya's own reduce function, anyway. For doing stuff by hand in Maya, the "draw reduce" tool that comes with the bonus game package is the mutt's nuts.

12 December 2003, 05:40 PM
I heard that the poly counts on the Dungeon seige models were around 800. I think I might use that for my RTS game because when I finally finish it that should probably be average graphics for an RTS in the future.

Any one know some mapping tutorials for Lightwave?

12 December 2003, 05:07 AM
Originally posted by Jonny Bubonic
If you use Maya, you might find this useful for reducing a model . It lets you assign "weights" to areas of importance, so that when you reduce the model down you can keep specific areas of detail that you don't want to lose. It's far better than Maya's own reduce function, anyway. For doing stuff by hand in Maya, the "draw reduce" tool that comes with the bonus game package is the mutt's nuts.

oh that Progressive mesh tool looks really impressive. thanks for the link. but this "draw reduce" tool what is it ( i am not a Maya user atm but am thinking of getting maya pretty soon) and where do you get the "bonus game package"?

12 December 2003, 07:18 AM
Here's a very good photoshop filter (and many other useful tools) to convert height maps into normal maps from nvidia:

nv tools (

My two cents: it's all about the engine efficency and the type of rendering (complexity of vertex/pixel shaders), but as a general rule remember that to keep 60fps, a radeon9600pro or gf4 ti 4200 class machine can't go beyond 40/50k triangles per frame.

12 December 2003, 11:25 AM
Draw reduce rocks ... You can get the bonus game pack in the download section of Alias` HP at, yep you guessed it, ... :)


CGTalk Moderation
01 January 2006, 11:00 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.