When in doubt, Weld (Large Pictures)
polycounter lvl 10
Evening guys, I bought TF2 a little over a month ago (I know, extremely late to the party) and fell in love with the exceptionally realised stylisation they managed to achieve. So anyway a good friend of mine (Evil_Ice/Daniel Bentz) agreed to lend a hand with all aspects of the process aswell as doing the animating/compiling himself. The main goal was to brainstorm a concept specifically suited for the game itself which i must say did not come naturally (having never painted one myself and being used to modelling from references). There is no name for this yet hence the random thread title, it will come soon though :poly121:
I do not own a WACOM pad so it had to be painted by mouse in Photoshop (very tedious), anywho let me know what you think and I'll keep this thread updated with progress as it is made
tl;dr?: First time producing/modelling my own concept. Need to buy a wacom.
Concept
High-Poly Mesh
Low-Poly Mesh
Ingame Texture Tests
Ingame Idle Pose
I do not own a WACOM pad so it had to be painted by mouse in Photoshop (very tedious), anywho let me know what you think and I'll keep this thread updated with progress as it is made
tl;dr?: First time producing/modelling my own concept. Need to buy a wacom.
Concept
High-Poly Mesh
Low-Poly Mesh
Ingame Texture Tests
Ingame Idle Pose
Replies
Still needs some optimization but looks cool like Jason pointed out. Make sure you're counting tris not polygons.
Also I'm not sure why the tape sags in toward the valley?
Wouldn't it stretch tight across due to tension?
Also the tape looks kind of weak to hold the bottle on.
What exactly was the idea behind this weapon ?
As others say this could be optimised alot, personally I think 50% the current complexity should be okay. On another note, the diffuse for the metal parts illustrated in yellow looks too much like modelling clay you would find in a sculpting application. It needs some variation and specular quality to emphasize the point that it is infact metal.
Yeah yeah, monitor can't display more than 1024x768, trust me, it irritates me far more than it does you ><
Just some little tidbit stills from animations in progress, to perhaps hint at some ideas underway. Will probably make some videos when they are finished.
In Kimono's defense, the mesh actually cannot be optimized a great deal more without compromising the the smoothing group(s). To achieve the best results in the Source engine, you want to keep as much of the low poly on a single smoothing group as you can get, and smoothing errors, depending on the severity, DO show through the normal maps.
We will be making optimized LODs for the world models of course, this low poly is for the view model, which is actually still more efficient than some of the stock TF2 models, such as the minigun or ambassadeur. Running ingame performance tests also reveals that with the average steam-hardware-survey user there won't be a significant frame drop in FPS when using the wrench/welder vs the other stock models, compared side by side.
You can achieve the exact same result and shave 2000+ tris out of that model and no one would notice. Just because there are heavier weapons in the game doesn't mean they aren't built with efficiency in mind. You have to keep in mind that part of the eng's overall weapon budget includes the buildibles. Where the heavy weapon is all that character carries. Since the eng is highly likely to see all of the buildibles and this weapon on the screen at the same time, it might be good to give it another optimization pass.
Also make sure you're counting triangles not polygons. Your counter looks like its counting polys, there is easily 2x that in triangles if that's the case. make sure you're not comparing your poly budget to the games tri counts.
I wouldn't mind taking a crack at optimizing it if you guys are busy.
Mark.N: The two gauges are mirror mapped.
Vig: It's most definitely counting triangles, I actually had to change the setting because it was telling me 4500 polygons. I will try and make sure that I find the time to take another look at the model before calling my part done, it was simply after doing some tests in the source engine everything seemed to benefit greatly from the extra loops added, perhaps it would seem I went slightly overboard :poly142:
Either way, it's still a pretty sick mod
As for its use...maybe it could be used to repair turrets? I would love to play with this thing... looks great.
Again not to say he couldn't shave off 1000 tris from this if he really tried, but saying he can cut it in half is just foolish, because he would lose what is so cool about it, the very much fitting, round, bubbly cartoon silhouette that the game is known for.
However i think there are some specific areas, like the beveled edges on the sides of the bolt type things(the 8 sided bolt that transitions into that smooth shape, and the thing below it, for example), since these shapes are round already, you shouldn't really need extra supporting geometry here. The bevels at the ends of these shapes are probably needed tho.
Now as far as source, what is wrong with using more hard edges(with uvs split accordingly so your bake comes out correct)? It also may be worth it to try baking in something other than max, like XN or maya to see if the results are any cleaner in source.
Optimization: Again, we are fully aware that the low poly COULD be lower poly, but this would disrupt the smoothing groups and the silhouette, and would not be worth losing the gorgeously smooth and high poly appearance we have managed to squeeze in under 10k triangles. Our greater fear, and this was tested thoroughly prior to achieving our current result, is when optimizing out some of the unneeded edgeloops, the smoothing errors become too greatly exaggerated and show through the normal maps, and cause hideous errors on the mesh.
(This here http://ice.annihilation.info/TF2/weld_ctf_2fort0000.jpg demonstrates some of the LEAST noticable errors we encountered using a more optimized low poly mesh)
The engineer does have buildables, but this does not mean that he has any better chance of looking at them all on screen at once, in fact usually he is only looking at his sentrygun, so he has LESS of a chance of seeing them all on screen at once, compared to a Heavy, Pyro, or Spy, who is avidly using his teleporters, and dispensers, while attempting to guard them. With that said, I'm sure if you'd like to try to compare the results, we can get you hooked up with the low and high poly meshes to try your own hand at it, but I've already done a fair amount of expirementation with the results with a few other models I produced for the OB Source engine for TF2 in particular.
(http://ice.annihilation.info/TF2/SkullBuster.jpg - http://ice.annihilation.info/TF2/SkullBuster/ctf_2fort0003.jpg)
While these are all good points, and we both greatly value optimization, we are attempting to create this as more of an artistic piece than efficiency piece, as we have both done primarily in the past. While compared to the other weapon models in the game, it is not excessive in triangle count, it is comparably soft and smooth, with a gorgeously rounded silhouette, and achieving that under 10k triangles is something I'm satisfied with.
The Normal Maps: just a quick edit here, we baked over 30 different versions of the normals using maya, max, and xnormal, and out of all of them, XNormal came out the best surprisingly (and fastest), and that is what you see here in the images, are the bakes created with XNormal. I believe EQ makes a great point here as well, that the high poly appearance really is part of what makes this a TF2 replacement.
Dials/Gauges: We are quite aware of the nonsensical and unnecessary appearance of the gauges, the repetitive nature, and the fact they say "Temperature" instead of "Pressure". This was actually by design, as in earlier images we had a far more realistic gauge faces, and decided they did not suit the stylized and cartoonish look of TF2 well enough. Upon even closer examination you may also notice that the dial-faces say "Temperature" on the top, and "PSI?" on the bottom, intended as a bit of a joke due to the fact that we decided that the colorful and illustrated temperature gradient looked better than anything we could come up with for Pressure. While it is true we could split the two gauges apart, and skin them both seperately, the real question that comes to mind is would it really be worth it? Apart from the more advanced cg art communities such as this one, and 'realism nazis' that won't be playing TF2 anyway, no one will notice them, as I have been animating the needles to spin rather wildly in action, and even oscillate at an idle stance. Again, not even remotely realistic, but realism is just too boring these days.
The Use: this is designed to replace the default engineer Wrench, which he uses to repair, build, upgrade, and kill! It's an interesting system you may be familiar with if you have played TFC or TF2, but you literally attack your buildables with the wrench to build, upgrade, or repair them, and it can still be used as a melee weapon to defend yourself along with your shotgun and pistol. Unless I write a serverside mod to allow this mod to ignite enemies on-hit, then it will not burn enemies or melt weapons, but it will still deliver a pretty nasty hit (same as the wrench) and will still operate identically to the wrench, with all the visual and audio changes, to make it feel more like a welding tool!
Thanks again for all the supportive and constructive criticisms and comments, and feel free to PM one of us if you need any particular details, or would like to contribute to it's construction. I'll try to post up some more updated previews of animations as things move along today, although I've been pretty busy all day thus far.
The tricount isn't an issue here. If you take a look at the default TF2 weapons you will notice that most of them aren't using normal maps, and are instead using edge bevels for everything. Using both in this case isn't that overkill I guess.
Knowing now that they are counting tris, the budget doesn't seem that unreasonable. If it was double like I was suspecting I think they should give it another pass.
EQ, I'm pretty sure hard edges cause seams. Breaking the big soft dull spec highlights.
As for functionality I think this is a spy-be-gone weapon so they don't need a pyro baby sitting the engy all game long.
Edit: (adding this back in so the answers make sense)
I have some questions about some areas, I know this is going to sound like a hardcore poly chopping whine but it looks like you guys tested this pretty thoroughly and I'm wondering why some decisions where made.
1) The ribs, do these carry around to the dark side of the model? I think its pretty standard practice for Valve to delete unseen faces from their View weapon models. Also these seem pretty dense could you keep a few on the edge that effect the silhouette and optimize the rest leaving the texture to do the work?
2) This seems like a very slight bend but with a lot of segments.
3) These holes are modeled? They are really small and I doubt you would ever see all the way through them. They look diffuse only in screen shots. As a matter of personal taste I would scale this object up given its some kind of control and the engineer has giant sausage fingers. Then maybe the holes would be warranted?
4) I'm not really sure why the tape has double the number of segments to go around the cylinders, it seems that one edge on the tape per edge its following would be enough?
5) Are those modeled threads at the base of the shaft by the bolt?
6) This just seems oddly high poly. There are a lot more segments here to round this out and flesh out very small inner details but the bottle and the tank are blocky (in comparison). You could probably start to remove every other edge as the details get smaller? It seems like its being kept all quads for quads sake?
7) You already answered this one, you're counting tris, knowing that its not as big of a deal to optimize and probably much easier left the way it is. Kind of a wierd fuzzy line though. "If its not hurting anything why not leave it in, but if it isn't helping why not take it out". Catch 22, and I guess it falls to whatever is easier to work with at that point.
To whom may be curious:
1) It is correct that backface optimization can be done, but not until animations are finished. This means that we may or may not know in advance which faces won't be seen during the process of creating and tweaking animations until they look right ingame. This does also mean that we may be wasting some UVW space by mapping areas that will later be optimized out, but there's too much guesswork involved in trying to accomplish this before animations are completed, which they are not quite yet.
2) The extra edgeloops on the curly straw may appear to be unnecessary, HOWEVER, they benefit more than simply geometry. If you have done some animation and rigging before, then you can probably understand why additional loops would be beneficial for weighting vertices in this region, to allow the use of fewer bones to control a larger area of the mesh. I've rigged this mesh so that the bottle AND surrounding tape, cork, valve-key coming out of the cork, and sections of the curly straw are all rigged up to separate controllers, using vertice weights to allow for manageable "organic" flexibility. This allows me to create some more dynamic, subtle detail-ridden animations that absolutely fit the last piece of the puzzle for this asset.
3) The holes on the Valve-key on the Cork there are 5-6 sided, and shouldn't really be hurting much. True they could just be normal mapped on, but I just NEVER seem to get the right kind of appearance trying to normal map a hole all the way through a shape, compared to just giving it a low poly hole in the middle and normal mapping around it. This is more personal preference, and I do believe one shared by Ben.
4) The extra edges on the tape around the beer bottle and canister have very similar reasoning here to #2, although I think Ben's low poly render there may have accidentally left some sub-d on, because as I recall the tape was modelled FROM the two cylinders of the beer bottle and the canister, making it strange that it would have more edges, so I'll have to look into it. Again similar to #2, the extra vertices do provide a great deal of assistance in weighted rigging the tape and the bottle to a single controller, and they animate quite well as a result. If the bottle/tape were to be rigidly fixed during animations, then we would have modelled the tape tighter, with no sagging in the middle, and less edges would have been far better, as they wouldn't ever run the risk of clipping the objects they surround.
5) No threading here boss.
6) The face of the gauge/dial is one place I agree %100, and I actually mentioned to Ben a few days ago. The silhouette is important here, to keep it nice and round, but I'm not sure it was his intention to leave as much modelled-in detail on the inside of the shape here as he did. This area could definitely use some optimization, and just let the normal maps handle it all.
Also, I completely agree that this would be a great additional as the Engineer's wrench unlockable, with some balancing tweaks here and there, it could do slightly less melee damage, but compensate by activating an after-burn similar to the flamethrower, which would help engineers AWARE of a spies presence to rid them. Couple that effect with the already planned %200 build speed and inability to upgrade built items, and I think it sounds like a great unlockable.
OK, breaks over, time for me to get back to work on the animations and compilation of this thing. (And since Ben is being a SLACCCKKKKER today, I guess I'm doing the LODs for the world model as well) Make sure you guys make fun of his top-hat and monocle wearing, tea drinking ways.
It looks good, though, man!! Love seeing tf2 stuff on here. It's the only game I play anymore
(and probably will be the only game I play up until Mass effect 2!)
http://ice.annihilation.info/TF2/Welder/ctf_2fort0012.jpg
http://ice.annihilation.info/TF2/Welder/ctf_2fort0013.jpg
http://ice.annihilation.info/TF2/Welder/ctf_2fort0014.jpg
http://ice.annihilation.info/TF2/Welder/ctf_2fort0015.jpg
Got a preliminary set of animations and sounds in place, I'll be refining both tomorrow, assuming that slacker Brit is back in a contributing mood we'll start knocking out the world model LODs as well, and the whole thing will hopefully be wrapped up by the end of the week.
Cool stuff man. Keep up the good work.
I also did more work on some tweaks here and there, and spent a few HOURS trying to figure out how to convert ingame demos into shareable internet media... so far I've been utterly disappointed, defeated, and disgusted...
And then I ended up with this hideous abomination, a truely sad excuse for a demonstration of this otherwise quite beautiful asset ingame:
http://ice.annihilation.info/Video/WelderTest.avi
or for about the same lackluster video quality, you can youtube it here:
[ame]
Let me take the time here to clarify that neither the animations nor the sounds are near completion, just some rough bases to act as placeholders while we do some general conceptual theorizing, to figure out what is going to look best ingame, and provide some minute iota of functionality to this pretty bizarre asset.
Yeah, the Source Engine doesn't do normal maps well. It might be better from when I used it last, but even still.
But I do agree some of that detail is a little nuts for a lowpoly model. I know it's hard to deal with compression and jagginess, but that's how things go. Although, when your doing it on your own, you can do whatever you want, but realistically it's quite high.
But there are many factors that can play in this as well, like LOD's, where the budget is spent (characters/weapons) or environments or whatever.
Edit: Your first LOD looks good to me, that's what your game model should start at imo.
And as if we haven't beaten this dead horse enough, my curiosity got the better of me.
I decompiled some of the View weapons to check them out here is what I found:
Engineer Wrench: 1900 tris
Engineer PDA: 1744 tris
Heavy MachineGun: 6196 tris
Sniper Rifle: 4256 tris
Spy watch: 1629 tris
Spy Sapper: 3181 tris
Spy Revolver: 6660 tris
HeavyFists: 5244 tris
Medic Medigun: 3129 tris
All of the models had unseen faces deleted, and a few looked like they where re-unwrapped or at least rearranged after they where optimized. The bevels I saw where pretty big and looked to be there to aid the shiloette and not smoothing so much. Granted most of their models are made up of primative shapes and not a lot of inner connecting pieces like the torch.
So I think the highest level of the torch is a hair higher than spec, but not by much. The 2nd LOD looks a bit low for the amount of detail especially if you take into account removing unseen faces.
I think a compareable weapon is the spy sapper 3181 tris. But really its not over by that much as it is...
Then you have to consider alot of these models may be combined on screen at the same time, such as the spy revolver, AND watch at the same time, and to top it off the two alternate spy watches have huge triangle counts. If that wasn't bad enough, you oughta check out of some of the other triangle counts (Ambassaduer), just imagine what THAT is doing to the average user who aren't even using custom models? Surprisingly little compared to the lengths we are going through to provide extra efficiency here with LODs at the very least.
Yes Valve has deleted backfaces on the DEFAULT view models, which would otherwise be ungodly triangle counts over 20k for many, but one very very important factor to keep in mind here is that Valve is NOT using normal maps for the default meshes, but relying only on geometry with the phong shaders to provide the smooth look. Currently we are employing a hybrid method of using normal maps to further round out details that the geometry alone cannot, and the extra edgeloops are necessary to keep the base smoothing intact enough to permit the use of normal maps.
Again, the proof is more in the performance, which currently is working BETTER on the Welder, than some of the default weapons, especially those like the shotgun, flamethrower, rockets and grenades, which utilize ADDITIONAL particle systems to provide some of their visual effects, which end up being more taxing more often than the relatively static effect of the slightly higher triangle count of the mesh we are using.
I've also been running constant performance, memory, framerate, and cache checks ingame with the welder, and have yet to notice anything take a hit more than the PDA, which coincidentally enough tends to cause me more performance drops when switching to it, due to the complimentary GUI system used by both the building and destroying PDA models.
Anyway, my point is, neither of us are ignorant to efficiency here, take my word for it when I say I've made my share of sub 1k triangle models with double digit tri count LODs.
Valve official TF2 models with high triangle counts taken directly from the meshes, these do NOT have LODs either:
Cloak and Dagger - [c_leather_watch.mdl] - 2494 (is seen on screen with another view model)
Dead Ringer - [c_pocket_watch.mdl] - 8613 (is seen on screen with another view model)
Ambassador = [c_ambassador] - 9439 (is seen on screen with another view model)
Huntsman - [c_bow.mdl] - 6564 (includes one arrow)
This just sounds outright ridiculous, but feel free to verify for yourself, and again, keep in mind that these were created with no LODs, and are used for the world model AND the view models... if these are making the official cut these days, and some in COMBINATION (ex Dead Ringer + Ambassador = 18052 triangles BEFORE the hands mesh!) should we really be holding on so desperately to efficiency budgets from 3-4 years ago? Won't there also come a day when 20k triangles will be considered low poly? I know were are beating a dead horse as you say, but my beliefs rest firmly with the results, and I'd rather compare a NEW art asset with the most recent official additions made by Valve software, than with models that were created back in the game's beta production.
[edit: just an edit to clarify, I'm not trying to be a stubborn mule here, so much as I'm enjoying the debate]
Has it been 3-4 years already? Consider it would have been in dev a lot longer we're looking at well before dual or quad cores. So the tri pushing power has gone through the roof since then. You need to up res it =P
I wonder if they would give the newer models another pass if they ported them to the consoles?
One test I would like to see is in the middle of a fire fight when a engineers base is under attack, especially if there are more buildings and maybe a pyro helping to defend, people coming out of teles, toss a heavy and a medic on ubber and all the sparking decals and you've got a nice little stress test.
I think that the additional optimization is in fact one of the reasons why valve has NOT ported the unlockables to the console versions of TF2. It would also mean going back and redoing their c_ model system, which would be laughable at best on a console like the xbox360. It would never be able to support the kind of high poly world models the way PCs currently can.