View Full Version : Apache AH-64D
02 February 2011, 09:12 PM
Alright guys, I just wanted to post this thread as this is a piece of work I have been recently working on for my portfolio maybe. I have created this apache within 3ds Max and I am looking for some constructive criticism really, I hope to import it into the unreal engine soon as just wanted to show it as a game asset.
Cheers for your time!
02 February 2011, 09:19 PM
Here is the Textured model :)
02 February 2011, 06:33 PM
tell us about this job by numbers ! how many polygon, its have Normal and specular maps ?
02 February 2011, 03:52 PM
Thanks Osama, Well as it stands the overall apache is roughly 14,000 polys however around 22,000 tris. The apache does have a normal map which i created through crazy bump and further editing within Photoshop. The same is said for the spec map too. I understand that I would of achieved a much better and more accurate result using Zbrush however I am not so confident using it yet so therefore I have tried to gain the best result without it.
02 February 2011, 11:24 AM
Could you give us a little more info..
What was your target platform?
by looking at the model, there are lots of triangles you can cull, and not lose any definition to the silhouette of the object..
02 February 2011, 12:52 AM
What other info may you be interested in, my target platform, I would have to say that this asset was created in mind of a game such as battlefield 2. Iam still trying to dig up more relevant info with regards to current game poly limits, I understand that Battlefield 2 uses around 15000 polys per vehicle.
With regards to the tris on my apache, were would you recommend revising over and in your opinion most obvious to the eye.
02 February 2011, 09:43 AM
the Target platform would be PC then? -
if it's a multiplatform game, there'd be a good amount of asset-sharing (to cut development cost)
I think for something like Battlefield, there'd be a higher-res top-LOD for the PC version, then lower-res versions for PS3 and X360.
I've done a quick paintover, showing the areas where I feel you can save a good number of triangles.
If it's genuinely LOW-res, and an in-game model, 22k is simply too high for this kind of model.
When you factor in LODs for a model like this (wich each LOD level losing half its triangles), you'll have an asset that weighs nearly 40k triangles.
The general rule for low-poly, is that if geometry doesn't contribute to the visible Silhouette as you move the object around in the viewport, then you don't need that geometry. (there might be exceptions.. like cutting faces to aid your UVunwrap.
As you model more and more, your instinct for what should stay/go will become far better. You'll be able to anticipate which triangles/edges can go, beause the texture will describe that feature just as well.
On the pic, I've traced the unecessary edges I saw instantly - there may be far more, but these are the obvious ones. The pink area is a little indistinct as far as form goes, so maybe it's completely flat, and you can save loads of triangles there.
Your rungs/handles: Really the only reason to stitch features into geometry is for local dynamic lighting (or lightmapping). Lighting solutions deal with stitched geometry far better than clipped - and there are fewer irregularities. But because these features are so small on your model, you can get away with clipping them through the Apache's outer skin. (you're gonna use Global illumination or final gather to render this model for your folio anyway, and raytracing your shadows is FAR more accurate than any in-game lighting solution you'll be using in a realtime engine)
Overall I think the model's good, and looks fine - but there's maybe twice as many triangles that is necessary. if you're on a project and someone asks you to model X in 10,000 triangles, the fewer you can do it in, the better. NOONE should complain if you achieve the same visual quality in 6,000 triangles.
*I think*, you should aim to achieve the same visual quality in about 7,000 triangles, using the same texture setup. I'm pretty confident that you'd get a very similar result, visually.
02 February 2011, 10:25 AM
Not to bad a model and I agree with Phunky_nick though about the poly count a ton of loops can be removed.
A few things small modelling things I have noticed though:
Tail rotor: your rotor is a little small and you have a cross with the blades at 90 degrees to each other. the longbows tail rotor is a bit odd see the picture for clarification.
Also I cant see both sides but I was wondering if you picked up on something where alot of people make a mistake when modelling the sides of the fuselage. as the AH-64D has a modified and enlarged left hand side compared to the AH-64a which is more of less symmetrical. Like I said I can't see but thought I'd point it out anyway. :)
Sure you got it but: http://www.primeportal.net is always my first stop for all pictures military
02 February 2011, 10:31 PM
Thank you for your help Nick and Blank itís much appreciated. I will definitely review the topology and hope to bring the poly countdown to something more reasonable. I will correct the rear propeller rotor too however I noticed you pointed out one side of the Apaches exterior is larger than the other. I think I took a short cut and modelled them the same so I could use the same UV template :p
One last question nick, it might sound daft however, what is LOD short for and what do you mean by clipping parts into the surface geometry.
02 February 2011, 10:45 PM
LOD means Level of Detail.
Pretty much every model made for a game-engine has to have a LOD, these days.
You get a Top-LOD: the highest detail-level, with the most triangles, and the keenest modelling. that gets rendered by the game engine when you're right up close to it.
When you're a certain distance away, it gets replaced for LOD1 - which is Lower detail (75-50% the triangles) , but from that particular distance looks the same.
if you move away that distance again, it'll get knocked down to LOD2 (50% fewer triangles again)..
It's a typical optimisation measure, so that the engine doesn't have to always render all those triangles.
If the LODs are properly implemented you can't notice - however, Games are made to a budget and a schedule and corners are cut, so you can occasionally see the LODs popping in.
Clipping: Just passing the geometry through a surface, rather than stitching it in.
There are geometry savings to be made with clipping, but problems if you do it in an uncontrolled manner. So with those rungs, you should be absolutely fine, with bigger pieces, it has an adverse effect on rendering performance (Overdraw).
02 February 2011, 04:59 PM
Well thank you for clearing that up, it certainly makes sense now. I guess I have witness what you mean with examples of LOD changes within game, for instance when the hardware sometimes catches up with its self.
As for clipping, I was unsure if certain objects have to be welded / stitched to the geometry if itís one complete object. I understand however in a games engine you can just place static meshes however you want without obviously welding / stitching.
I have started reviewing the topology on my apache model and was wondering if you could lend me one last piece of advice. When creating model topology I understand the less tris...the better, I usually try and model with an equal 4 sided polys when I can. So the question I ask is, is it better for instance to have an 8 sided poly, or is it better to connect the verts with the 8 sided poly and consequently have more tris. I hope that made sense :/
02 February 2011, 05:09 PM
I'm not sure I understand what you mean with "when hardware catches up with itself".. the different LODs cut in at fixed distances.. The only time you'll notice is when a LOD is drastically different to the one before it.
Modelling in quads or tris is about comfort, these days - comfort in the 3D software . (it used to be about getting artwork to run optimally on weaker hardware... PS1, N64 etc.. )
I don't see how you'd have a situation where you'd need to make allowances for an 8-sided polygon, though. ?
02 February 2011, 05:46 PM
Maybe I have misunderstood, I recall on some older games such one I remember g.t.a vice city, it sometimes had that effect were you could enter a room or drive across the map at a fast rate and the environments would appear to have much lower texture resolutions for a few seconds before returning to its original texture res.
I canít say I would try and model topology with having more than quads: p however is it acceptable to cap a segment up to more than 5 sides? Or is it best to cut it into smaller segments
02 February 2011, 02:46 PM
I'm not too sure about Vice city's rendering,
as it was PS2 (and fairly late in the PS2's life), I'd imagine that some mid-far distance rendering quality was sacrificed for increased quality in nearby assets - but that's me speculating, I absolutely don't know for sure.
I can't say I fully understood your last line here, but i'd say if you're working neatly enough, there should always be an alternative to having 5-sided polygons..
(sorry for the late reply!)
02 February 2011, 02:46 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.