Important things to know when creating graphics for PV3D

30 01 2009

Behind this blog a lot of activity is going on, assetswise. The best thing with developing a game by yourself is that you can actually put something aside for a while and do something completely different. Tired of coding? Well grab that keyboard and create some music then? No, what about modelling? Not today, huh? Well, there’s always, texturing, level design, 2D-art, sound effects, visual effects, PR, in game dialogue, websites… etc etc.
So I’m creating a lot of content for the game and want to share some fundamental tips when it comes to modelling and texturing for PV3D.
First we need to realize that PV3D is way behind when it comes to 3D performance. We all probably know the reasons why and the fact that it is a great new experience for the flash environment doesn’t make it fresh and new as a 3D experience. Even back when 3dfx released it’s first Vodoo chip more than 13 years ago the computer could render polygons much faster and more accurate than PV3D is able to do. Still this doesn’t have to be something genuinly bad. Personally I see this as a wonderful creative challenge to actually be able to trick and cheat my way through PV3D to push the limits so the endresult still can intruige, shock and amaze users.

There are several PV3D users out there that compare the whole situation with the old demo scene from the C64, Atari and Amiga eras and I agree.

Now, then. Question is what CAN PV3D handle and how do we create optimized assets for the engine?

Let’s start with modelling:

Every triangle counts! Its obvious that it does but still people are really bad modelling optimized 3Dmodels. Even if you think you cannot remove any more polygons without distorting the whole model, you probably can. Let’s take my car as an example:

defender This model was created for another engine and even though it runs smooth as it is now on our testscenes , the model is way too detailed both for the game and for PV3D. At the moment the jeep is built up with 1108 triangles and a lot of them are very very useless. We always need to consider how close the camera will come to the unit as it’s closest and calculate whether some details are worth it.  The two crash pipes on top themselves use way too many polys and could easily be built up with 2 or 3 planes facing upwards. As the game will be a top viewed game I can remove all polygons underneath. There will be no physics in this game so the jeep won’t roll around showcasing whats underneath.

Rounded areas is always a problem so even if we cannot have square wheels we could easily take a few of the polys there. Remember that the camera will be rather zoomed out most of the time.

You can always keep the detailed model as well as might want to try a LOD technique as the camera moves in. LOD (Level of detail) is working in the way that when the camera gets closer to an object, the object switches to a more detailed model whenever the details will be more visible. In this manner you can save a lot of resources when the camera is zoomed out. I think I will try this technique in the game. Also remember to check for loose vertexes and faces when you are creating your model. In several 3D-apps there are a few automatic optimizefunctions that both removes unused vertices and also tries to optimize the model (based on angles). You could come very far with these features.

I cannot stress enough how important it is to keep the polycount down. There is nothing worse than realize in the middle of your development you will only be able to spawn 3 models and 2 bullets at the same time in a shoot’em up or the gameplay will be ruined by a framerate at 6 fps. Even though it runs smooth when you test your models, it might not do that when there are 3 in there + AI code + 2Dfx + HUD + networking etc etc. A lot of games has gone down the drain just because it lagged too much and the joy of creating the game just disappears.

Now there is another problem with PV3D. As PV3D is not able to set perspective correctly on each face, it “fakes” perspective by “skewing” the triangles. It works like a charm when a lot of triangles are involved and when its facing the camera but if you eg create a plane  with no division (only 2 triangles) and then rotate it on the y-axis you will clearly see a great distortion of the texture when ever it is facing any of the sides. This is not happening in a normal 3D-engine but is a terrible must in PV3D due to what we are given through Flash. This means that certain flat surfaces that easily could be built up with 2 triangles not instead requires a lot of polygons not to distort terribly.

This will be a huge problem for the arena ground in my game so I will think about a solution.

What about texturing then??? Well, how you handle textures in PV3D are more forgiving than triangles but if you try using a texture (BitmapMaterial) compared to a ColourMaterial you will of course see a huge difference in performance. But still.. we want textures, right?


Above you can see the texture for the defender cannon. The difference between the 3 is the size of the texture (from left to right 128×128 , 512×512 , 256×256). Comparing the left and the right cannon you can see that the texture is of course smoothed out with 128×128 and there are more details in the 512×512 one. Same here is the question, how CLOSE will the camera get to the object? At the same time you must ask yourself, is it worth it?

I know that in this game during gameplay, the camera will be very much zoomed out just like the pic below:

hudSo in this case, any big textures are just a waste om memory and processor resources that I could use for a lot of other things. Still, usually in PV3D projects, I see a LOT of too small textures giving you that squary pixel textures. That aint cool! Using a 256×256  or 512×512 texture is more a question of memory than really that much processor so is you are seeing too many and too big pixels, try a bigger texture.

The classic way of deciding which size the texture should have is to make it on any size of 32,64,128,256 or 512. Now, (correct me if I’m wrong), believe that is a heritance from how the arcitechture of the gpu’s were built and one were always recommended not to use any bigger than 512×512. I do not believe that’s the case for PV3D as there is no GPU Hardware acceleration involved here so we’re using the normal memory. Still, going over 512×512 shouldnt be necessary unless you are working with big environments (eg the Arena ground in this game).

This texture-tips has only been about the “diffuse-maps” (the colour that are visible on the model) there are a lot to be covered when it comes to textures and materials that I probably will bring up later on in this blog. I’m not sure if any of the shaders will be used in the game but I would still like to mention them both for experience purposes and to uphold this blog as a tutorial blog :)

Now go and create some assets! Here’s some wip of mine:



Paint the paper

28 01 2009

My guess is that PV3D chose the word “paper” because the way to treat 3D objects is very much like folding paper in different ways. Even if a 3D-object in the end could look like a very solid rock it is built up with simple planes and triangles that is carefully bent and rotated (actually the building blocks gets smaller than that but let’s rest here for a while). The small sides (triangles) that are building up the world doesn’t even have two sides. If you are looking on the other side of a triangle you will actually see… nothing. Nothing at all! So to even create the simpliest sheet of paper you will need to create two planes and flip one and stick it to the other to make it look like in real life (have both a front and a back side).

Now, technically, Papervision have a lot to wish for when it comes to performance and no matter how amazing you might be codewise you won’t just be able to create detailed lifelike models because it will take you thousands and thousands of these triangles to fold and model the details of a reallife complex object. Here is where “texturing comes in”.

Instead of using all these resourcedemanding triangles to showcase something you can actually paint stuff right on the triangles. Say you want to create an orange. Instead of creating every single small bump on an orange you could just paint those bumps on an image and then ‘wrap’ this image around a more simple sphere still creating the illusion of an orange.

Now, I’ve started up creating a few of the models I want to use in my game. Below you will see my unit: “DEFENDER” – a smaller buggy-like car which later on will have a big ass cannon on the roof.


To create such a model, you will need some kind of a 3D-program. I’ve been working with 3D Studio MAX for the last 10 years so that is choice of app. First of all you create the actual “folding” of the object. Placing all triangles where they should be, bending rotating and modelling out all planes will give you a model something like the untextured one above. The texture in the picture is what I draw on a regular image in Photoshop. Wrapping that image on the planes causes the textured model which gives an impression of a much more detailed model than it really is. Good textures are very very important to create a good looking game. Spend a lot of time on that!

If you are interested in special tutorials in modelling, texturing, UVW-mapping etc.. just give me a call and I’ll see what I can do. At the moment I just want to get my “defender” to be a part of the game. To use these kind of custom models you will have to save them into some format that PV3D recognizes like *.3ds or collada.  In this case I will save my model into the *.3ds format and use the inbuilt parser Max3DS in PV3D.

	import controllers.*;
	import interfaces.*;
	import org.papervision3d.lights.PointLight3D;
	import org.papervision3d.objects.DisplayObject3D;
	import org.papervision3d.objects.parsers.Max3DS;
	import org.papervision3d.view.BasicView;
	import transformers.*;
	import units.*;

	public class View3D extends BasicView
		private var defender:AUnit
		private var defenderMesh:Max3DS
		private var light:PointLight3D

		public function View3D()

			// set the camera in a nice position
			camera.y = 200
			camera.z = -1000
			camera.zoom = 90

			// place a light somewhere (just to get used to it, this one doesn't do much at the moment)
			light = new PointLight3D()
			light.x = 1000
			light.y = 5000
			light.z = -6000



		private function startLoadMesh()
			// lets load our mesh
			defenderMesh = new Max3DS()

		private function meshLoaded(e:Event)
			// mesh has been loaded and is now put in as a parameter in this new object called Defender
			defender = new Defender(defenderMesh,new LocalController(),new CarTransform(7,0.85,0.3));

			// startRendering() must be called for the 3D-engine to actually start drawing each frame

			// each time the 3Dengine redraws the screen, onFrameTick is called.

		private function onFrameTick(e:Event)
			// lets transform the unit due to the control-data it recieves.

The code above actually loads my mesh and adds it on the scene in 3D. Not only that, the mesh is passed into my new class: which is extended from AUnit so now FINALLY the chain is connected and I got a unit that do have a mesh, a control component and a transform component. I got my first primitive unit driving around on the scene!! Well.. almost, I just need to create that Defender class..

package units
	import interfaces.*;
	import org.papervision3d.objects.DisplayObject3D;

	public class Defender extends AUnit
		public function Ball(mesh:DisplayObject3D,controller:IUnitController,transformer:ITransformer)



We need no more than that to create our unit. Remember now that we can create as many different components as we want to. Just combine them, reuse them and experiment with them to create new units.