Car modeling techniques
polycounter lvl 13
Hi I am after some advice on car modeling techniques, more specifically for games
in most games now damage is a big part of game play and thus how the car is modeled, this must mean that all the panels are now separate instead of backed or textured panel gaps.
as games require a limited polygon count how would one model this, would you model all the car as a single mesh, getting edge flow following everything through the entire car or model each part separately allowing different edge flows from piece to piece.
for example the lights on a car are not attached to the metal body, there may be a lot of complex geometry leading into the lights area, the lights wouldn't necessarily need all those edge loops to continue through them, so in industry would they model it separately disregarding the surrounding geometry or would the car be modeled as one and then detached at the end to its separate parts to ensure perfect match up of shape? (there could be minute differences in shape with different amounts of geometry on adjacent panels. also I would image that it would be entirely possible than mesh smooth may also change some shapes right?)
Edit*
I remembered this image from some time ago, this is one of the best "potential real time" car model I have seen,
they have theirs continuing over some parts, however the front bumper is different and the thing that most stood out to me was that it has tri's, now this inst anything bad, however for meshsmooth I was always taught to have 100% quads, if the car in this image was not intended for meshsmooth then there is a lot of optimisation to be done, especially at the top of the front wheel arch, the polygon distribution seems a little off, having too much in one area than another.
in most games now damage is a big part of game play and thus how the car is modeled, this must mean that all the panels are now separate instead of backed or textured panel gaps.
as games require a limited polygon count how would one model this, would you model all the car as a single mesh, getting edge flow following everything through the entire car or model each part separately allowing different edge flows from piece to piece.
for example the lights on a car are not attached to the metal body, there may be a lot of complex geometry leading into the lights area, the lights wouldn't necessarily need all those edge loops to continue through them, so in industry would they model it separately disregarding the surrounding geometry or would the car be modeled as one and then detached at the end to its separate parts to ensure perfect match up of shape? (there could be minute differences in shape with different amounts of geometry on adjacent panels. also I would image that it would be entirely possible than mesh smooth may also change some shapes right?)
Edit*
I remembered this image from some time ago, this is one of the best "potential real time" car model I have seen,
they have theirs continuing over some parts, however the front bumper is different and the thing that most stood out to me was that it has tri's, now this inst anything bad, however for meshsmooth I was always taught to have 100% quads, if the car in this image was not intended for meshsmooth then there is a lot of optimisation to be done, especially at the top of the front wheel arch, the polygon distribution seems a little off, having too much in one area than another.
Replies
Loads of different ways of doing things, Im no expert, but I prefer subDing a shell, with most of the lines in Sub D, then collapsing and deleting loops to optimise. I like to stay in quads for as long as possible, so do most of the lines in subd when i can. When it is a massive pain, Ill collapse the turbosmooth and do the weird lines after.
Headlights I sort of make up as I go, but I generally dont have them connected to the main body.
http://speedhunters.com/archive/2011/11/17/behind-the-scenes-creating-cars-need-for-speed-run.aspx
You are almost always going to have to use tris, cars are definitely not designed with polygons in mind, and you are most likely not going to be using normal maps, so you'll need a lot of tris in some areas just so it shades well.
Computron: Max, as far as I've used it, is notorious for shading everything like crap even when your flow is smooth. They even made a hotfix for 2012 in order to "mend" their wonky normals... Preview with xoliul or 3pointshader and you should get a proper result. Also, Maya shades things a lot better for what I remember.
Edit: Clicked the link after I commented and whaddayaknow, they talk about the Aventador
It makes a lot of sens that they baked a meshsmooth and then played around to get tri's and a bit of optimization, something I hadn't thought about because the intended purpose of this exercise I have undertaken was to see how UDK's tessellation worked with something like this, along with Lod's and the normal method of stuff like this.
as such I have tried extremely hard to keep it all quads in the hope that tessellation will work nicer with the model, while I have had a fair bit of experience with UDK, tessellation and the new features over the last year I have not had much time to explore and test.
the overall shape of the car didnt take too long to do, but details like the lights and some of the trims are now trippeling the time its taken to make it due to this quad problem, as such I have made a saved version and gone into tri's to attempt to see how tessellation and meshsmooth will work though I already have a small pinching effect with the mesh smooth.
As som1 already had posted Speehunters link
notice black artifacting on middle top left of hood, this is solved by adding more polys,tris to such sharp angles
you have a good start there but without more subdevisions then you are not going to get the rest of the detail on the hood.
http://udn.epicgames.com/Three/SettingUpVehicles.html
(i ll let u win this time, im busy with other stuff lol)
As a general case study lets take Forza for example, I would like to know about polycount in game and selection screen, shaders/materials, LODs how many, collision set up, rig set up and so on.
Feel free to have your best professional guess as I appreciate we may not be able to get official statistics.
Also I love this bake down by Simon Fuchs - http://www.simonfuchs.net/folio/
Koenigsegg CCX Render (HD)
http://www.gtplanet.net/looking-beyond-gran-turismo-5/
www.youtube.com/watch?v=5fIqRk6pYQ8&eurl=http://www.bestmotoring.cn/read.php?3308
GT5 Cars were about 100k total. That is inside and outside. So for the full car on the ps4/pc/x1 it would easily be double to three times that *if needed*. At some point diminishing returns is going to kick in and the smoothness will plateau for most cars, if it hasn't already.
Could I have a rough polycount for these games.
Project CARS
Forza Horizon 2
F1 2014
Wreckfest
Crew, The
Driveclub
GRID Autosport
Project CARS
Runs at 60fps on PS4, But Not on Xbox One
Each Car in Console Versions Has 60,000 Polygons
PlayStation 4 Handles Project CARS at 60fps, But Xbox One Currently Doesnt Cars Have 60K Polygon Count on Consoles
PC has 300K Polygon Count
"it has been confirmed that each drivable car in the console versions of the game has about 60,000 polygons. On the other hand, cars in the PC version of the game running on Ultra high settings feature polygons between 200,000 to 300,000, and the number comes down to 60,000 when the game is running on high settings. Project CARS comes with normal maps to cover up for low polygon count on consoles. If compared to Driveclub, Sonys upcoming racer has about 250,000 polygon count per car, but the game has been designed to keep the number low at the same time."
Read more: http://wccftech.com/project-cars-60fps-ps4-60000-polygons/#ixzz3eUTEKYjl
So are people modeling them subdivision ready and they are subdivided in runtime ? Wouldnt that make sense from a memory standpoint over saving out and importing the various versions as you had only one mesh at a time and save on the other LODs?
Looking at the first screen, it's clear they use custom vertex normals. So maybe they are degrading from a CAD model, or use something like meshfusion.
Does this mean in tri's so it would be 60k tris or would be 120k tris?