Help with super low poly modeling

jibberishballr

Hi everyone! I've been modeling for a while, mostly in Maya, and have recently started working in mobile game modeling. I am having some trouble and would like to know if anyone has a general workflow for modeling super low poly models(100-300 tris)

I am currently working on this model and have 100 tris to do it in!

Steve Schulze

I don't know that there are many real secrets to working on extremely low poly models.

In the case of that thing, make it segmented. Take advantage of alphas where you can. Texture on the slot on the side of the sphere. It's going to be pretty blocky no matter how you model it, but you should be able to make a passable model with some smart, efficient modelling.

Andreas

Use a geosphere for that sphere. I love me some geospheres.

Snader

Can't reasonably be done under 100 triangles.

With extruded tapered sides those sideshields are (6*4)+(2*2)=28 triangles per shield, 56 in total. That leaves only 44 triangles for the scope, sphere, antenna and that horizontal rod. There's a way if your engine supports double-sided faces but then (technically) it would go above 100 triangles again, or alphamaps but alpha is more expensive to render than a few dozen triangles. In short: this isn't really going to work unless you use tricks that cost more in performance than the few triangles you save.

I'd advise you to make a new concept, and focus on getting a design that's more possible within 100 triangles. (at least, I assume that is your concept?) Actually, how about this: if you can draw me a concept that you think you CAN make in 3D (without alpha or doublesided geo) - I will give you the model belonging to this silhouette:


(which, sadly, does use doublesided geo for the cross-shields)

D4V1DC

pay me?
I got an idea I'll try and see if it works if it does, pay me.

No seriously pay me.



j/k
I don't think my plan will work but It might.
Doing it now and see, then we can talk.
:poly124:

I'm getting the feeling of troll is trolling.:poly142:

wailingmonkey

fun challenge, but I agree with Snader...mine was 100 triangles with
only single-sided poly wings...add another 20 triangles to yer budget
and it's do-able.

D4V1DC

98 no double sided.

I think the wings are like the hospital sign +

Fun.

Edit:
More triangles would be great though.

Edit2:
72, no form but coolness.:poly142:

jibberishballr

I ended up getting it done with 72 tris but it was a pain in the arse...trying to convince the mods to give us a little more breathing room.

Andreas

Pics or it didnt happen :P

Tekoppar

This is what I got with 62 tris.



68 tris

WarrenM

You could spend a few more and round out that front piece juuust a little.

Tekoppar

WarrenMarshall wrote: »

You could spend a few more and round out that front piece juuust a little.

Well I figured since it was such a low tris count it would be viewed from a bird perspective, but I updated with a new one that's 68 tris. If he would tell us how this would be viewed it would be easier to help him.

Adam L. Gray

Well, here's me hopping the bandwagon then, 96 tris. It took me roghly 10'000 manly manhours.

ZacD

100 tris, including double sided.

SnowInChina

16 tris !

Fingus

12!

ZacD

1 tri

Adam L. Gray

Nice one fingus, yours is my favourite so far. But you could probably have optimized this one quite a bit by shaving off that edge here marked in red, and added a vert in as in the po.



I had another go at it myself too, 2 verts:

SnowInChina

you may win this one zacd, but beware of my return !

Snader

You could even do it as a single vertex. (if that vert is a vRay proxy.)

Anyway, Jibberish - what did you end up with?

jibberishballr

Snader wrote: »

You could even do it as a single vertex. (if that vert is a vRay proxy.)

Anyway, Jibberish - what did you end up with?

Forgot to post back on this: I ended up making it with 70 tris, and normal mapped it, which was still a pain. heres the geo, i have the textures lying around somewhere...

odium

Normal mapped it...? And yet stuck dead on not being over 100 tris? Hmm...

passerby

ya think you need to have a little word with you really out of touch tech guy if they want <100 polys but allow normalmaps, the extra texture call, and the math in the shader for a normalmap will cost way more than a other 100 or 2 polys.

odium

100 polies on the nose. I could have shaved off a fair few polygons on the main body, but the goal was 100 tris so I used them to smooth out the shape, which will become a lot more obvious when shaded/textured.



Download: http://www.team-blur-games.com/odium/thing_100polies.zip

Trouble is, the alpha testing and defom tube are likely more expensive than just using a few more tris, but both are certainly far, far, FAR more cheaper than using normal mapping. Normal mapping such a low poly asset is totally pointless anyway. I think whoever is your art lead, likely doesn't really "GET" this form of media.

Snader

Yeah. Your director or lead artist or whatever title he has is silly. Sub 100 triangle limit, but normalmaps are ok-to-go?


Just nab this model and add a simple diffuse texture. No alphamapping, normalmapping or whatever tricks required. Heck you could even go totally retro and use only unlit vertex colors and it'd still look passable.

jibberishballr

passerby wrote: »

ya think you need to have a little word with you really out of touch tech guy if they want <100 polys but allow normalmaps, the extra texture call, and the math in the shader for a normalmap will cost way more than a other 100 or 2 polys.

Can anybody give me some reference links to show to the directors? I'd like some evidence to support the claim that using a higher tri count will be more efficient than using normal maps.

Snader

Most hardware nowadays is more limited by fillrate than by vertex counts:
PS1 0.36 mil tris/sec, PS3 1200 mil tris/sec (3333 times as much)
PS1 33mpix/sec, PS3 13200 mpix/sec (400 times as much)

N64 0.15 mil tris/sec, Wii 50 mil tris/sec (333 times as much)
N64 31mpix/sec, Wii 950 mpix/sec (30 times as much)

Xbox 117 mil tris/sec, 360 1500 mil tris/sec (12 times as much)
Xbox 1866 mpix/sec, 360 8000 mpix/sec (4 times as much)
(data from http://www.neeyik.info/consolespecs )

So every console family shows an order of magnitude of difference between the increase of triangle counts. That means that you can now use about 10 times more 'extra' triangles when compared to texels as you could fifteen years ago.

That's assuming we still made models the same way we did then: simple diffuse map, and MAYBE a specular map or alphamap (lets say an average of 2 textures per model). We don't. Any decent material nowadays uses a diffuse, spec and normal, with often an alpha or glow map (lets say an average of 4 textures). That doubles the difference. Triangle count is 20x less likely to be the bottle neck than textures are.

But wait! There's more! Order now and get an extra exponential increase! Adding twice the polygons will make something twice as heavy. Making the resolution of something twice as high means a texture 4 times as big (twice as wide, twice as tall) so that means for every time you double both your texture resolution and your polygon count, the bottleneck shifts more and more and more towards textures.

And as a last bit, a vertex location is still used pretty much the same as back then, but a shader instruction for a normalmap is more costly (calculate several angles together, multiply by lighting) than an instruction for a diffuse pixel (multiply color X lighting) and this will undoubtedly make texels yet more costly - though I have no idea how much and am afraid to hang a number on it.

Some pullquotes/links from other places.

http://www.polycount.com/forum/showthread.php?t=50588 :

Polycount is not the huge issue people were rightly sure it was previously. The bigger issue now is texture resolution because all assets carry 3 textures as standard, normal map, diffuse and spec and thats before you have additional mask light textures for emmisives and reflection or whatever other stuff you are supporting in the shader.

Shader complexity is also a bigger issue now because it requires longer rendering time.
Section counts are a bigger issue , meshs carrying 2 of each texture and thus requiring a 2nd rendering pass.

http://wiki.unity3d.com/index.php?title=General_Performance_Tips :

The performance gained from using extremely low-poly meshes (like under 500 polygons) is minimal if at all there. The majority of graphics cards have hardware transform and lighting, which means they can process ridiculous amounts of polygons per second. Additionally there is overhead for submitting a mesh to the graphics card to render, so being really thrifty on polygons is probably only making your game look blocky.

This pertains to the amount of objects in the draw call vs the amount of triangles per draw call. http://www.derivative.ca/wiki/index.php?title=Optimize_Geometry_for_Rendering#Geometry_Batches :

The 'cost' of one of these draw commands is the same regardless of the complexity of the geometry. For example telling the GPU to render a single triangle has the same CPU cost as telling the GPU to render 10,000 triangles.

in other words, 10 objects of 1 tree cost more time to process than if you were to merge them into 1 object of 10 trees.





edit: this is all based on console/PC hardware, but similar trends will be happening on mobile devices. After all, it's mostly about miniaturizing technology and reducing the power draw.