Alrighty, I decided to get into making models for maps because I have a few maps that are using wayyy too much func_detail to create all these details in the maps.

On one of my maps (that I cannot mention), I needed some lamps to be attached to posts, so I made one, textured it, and exported it. It works wonders. The only issue I have is with the amount of polys/triangles a model can have. I try my best to make the lowest amount (said lamp had about 400 polys), but I'd like to know if anyone has a general idea of the amount of polys that stock models have. I've heard people creating models out of over 10000 polys and still being acceptable for DOD:S. I can't really judge on the performance part because my machine is pretty much top of the line right now and it's hard for me to see any slowdowns in my maps so I can optimize them :\ (show budget is a blessing though).

So, any ideas on the polycount of models? I'll also be releasing my models here for you guys to use :)

10,000 is acceptable for stuff like the Tiger tank model (default is 10k I believe). I know for a fact that some models can have up to 39k (with only ONE in the map, though), but I know that will probably slow down older systems. So basically, small props like lamps shouldn't really be above 1,000 and vehicles and stuff shouldn't really be more than 10k to be safe.

Sweet, so I guess I should continue modeling the way I do now ;)

Yer acceptable limits are something like this below

Vehicles - 8-12 thousand polys
Complex world models (Not furniture etc) - 1 - 8 thousand polys
Basic props - 1 - 2000 polys

Thats just a rough guide on what I think would be acceptable for most props etc

I ran into a problem with textures on one of my models (check the screenshot).

It's a fence with tiny supporters (the black cubic pieces) and each post is made of concrete. Basically, for some obscure reason, the post is acting as if it were translucent, and you can see everything behind it (look at the screenshot, you can see the barbed wire and the black supporters from the back of the post).

My texture has an alpha channel because my model has both the barbed wire and the posts in it. So basically, the barbed wire part has a little alpha, and pretty much the rest of the texture's alpha is white (so you can see it).

What's up with that? :\

Regarding polycount on models, Just use the HLMV and reference a similar model.

In the model texture .vmt you can try using "$alphatest 1" instead of $translucent 1". May work, or...for the post, you could just reference an existing concrete texture that came with HL2? So your model would have 2 materials, your wire texture and the HL2 concrete material, that you just reference.

I used this method frequently on custom models I created for Unreal 2004.

You can open models in HLMV and see the # of polys i think. That way you can compare with stock models.

you could just reference an existing concrete texture that came with HL2? So your model would have 2 materials, your wire texture and the HL2 concrete material, that you just reference.

Just be aware that this loads 2 textures into memory.

HLMV isn't an option because, for some odd reason, it crashes all the time no matter what version I use :\

Thanks for the references though, and the alphatest worked by the way :)

If you're worried about polycount I suggest you read the following on using and making LOD levels.

http://developer.valvesoftware.com/wiki/LOD_Models

I'd also like to know if anyone knows how effective models are at lightening a map's budget. I've heard in some threads that using models instead of detail brushes is actually more economic.

From what I understand, a complicated model has the same weight as a simple detail brush. So for the same buck, you can get a lot more bang.

Just don't overdo it with the number of models visible at the same time. A hundred tiny models still take a lot of performance, probably even more than 100 detail brushes, because Source has to make one rather expensive driver call per model. But(!) those 100 tiny models combined into one would be much faster than 100 detail brushes.

So is anyone capable of giving me a ratio of how many model triangles i can get per detail brush triangle?

I doubt anyone could give you a ratio, it depends on too many factors. Just do what Valve did and you're on the safe side. I just wanted to point out that Source tends to freak out if there are too many models visibible at once (no matter how many polygons they have). Can't even give you a number, depends on the graphic card. But say you want a fence and you model just one section (1 pole and the part between), then you build a big fence out of those small sections in the map. Then it would be cheaper if you model larger sections or the whole fence as one model.

I doubt anyone could give you a ratio, it depends on too many factors. Just do what Valve did and you're on the safe side. I just wanted to point out that Source tends to freak out if there are too many models visibible at once (no matter how many polygons they have). Can't even give you a number, depends on the graphic card. But say you want a fence and you model just one section (1 pole and the part between), then you build a big fence out of those small sections in the map. Then it would be cheaper if you model larger sections or the whole fence as one model.

That's interesting, but wouldn't a map with a zclip actually be better to use shorter pieces? I actually have that issue that I have a very big fence around my map and I was wondering if I should make it one big model or multiple small ones. Considering the Z-clip is at about 1750 units, smaller models would be more economical because not all of them are showing?

My map already shows over 100 models in some views, but I used the visibility spheres (or whatever they're called) to make them invisible after a certain distance as well...

Yes, I'd think so. If it's such a big fence then split it up into smaller sections.
Although I'm not sure how Source handles Z-Clipping, it might clip on a per poly basis. In that case it would still clip, even if it's one huge model. But I guess it just renders the whole model if a part of it is visible.

ranson i like this fighter plane crashed version, good job