Next Gen Tri count question
polycounter lvl 7
Hey Guys, currently working on an asset that I wanted to target for "something beyond this generation of games" or "high end pc" asset.
Originally this asset was started as a tiny prop to do some sculpting, but as I progressed I realized that for the amount of detail I was putting in, it would probably be a close up prop (maybe a cutscene, or something right in front of a player in a first person game). As a result, I topogunned (now a word) my low poly a bit higher than expected and was wondering if I went overboard. The image I was using as a base is this:
And my sculpt (to show detail level) would be here:
(there are other parts finished as well that are not shown in the sculpt, as well as an interior)
So here is the hard part to admit. The final tri count of the entire thing is currently 7.5k. A rough breakdown of how it is so high is here:
Main body ~ 1k
Metal clasps and stuff ~ 700
Lid, and wooden side pieces that hold the handle ~1.5k
Handle - ~ 4.4k
I am aware this is on the high side, and I wanted to do a tiny bit of explaining as to why so I don't have to later.
The cylinder of the body is quite high in division count as I figured I shouldn't be showing tessellation with it at around 1/5 of my screen
The handle, for texture space, is mirrored. That added a few tris
The handle as you can see in the original image, has some pretty complex detail. As a result I had to add supporting loops all over the place to get a decent bake without normal artifacting at the distance I wanted to prop to be shown at.
TBH most of the tri count on the handle comes from the silhouette, and the supporting loops. Anyway, if you guys have any thoughts that would be great! If needed I can post some wireframes, but I am looking for more general advice that specific ideas for cutting down (I can do those on my own, they would just sacrifice the final quality).
Thanks in advance!
Alex
Originally this asset was started as a tiny prop to do some sculpting, but as I progressed I realized that for the amount of detail I was putting in, it would probably be a close up prop (maybe a cutscene, or something right in front of a player in a first person game). As a result, I topogunned (now a word) my low poly a bit higher than expected and was wondering if I went overboard. The image I was using as a base is this:
And my sculpt (to show detail level) would be here:
(there are other parts finished as well that are not shown in the sculpt, as well as an interior)
So here is the hard part to admit. The final tri count of the entire thing is currently 7.5k. A rough breakdown of how it is so high is here:
Main body ~ 1k
Metal clasps and stuff ~ 700
Lid, and wooden side pieces that hold the handle ~1.5k
Handle - ~ 4.4k
I am aware this is on the high side, and I wanted to do a tiny bit of explaining as to why so I don't have to later.
The cylinder of the body is quite high in division count as I figured I shouldn't be showing tessellation with it at around 1/5 of my screen
The handle, for texture space, is mirrored. That added a few tris
The handle as you can see in the original image, has some pretty complex detail. As a result I had to add supporting loops all over the place to get a decent bake without normal artifacting at the distance I wanted to prop to be shown at.
TBH most of the tri count on the handle comes from the silhouette, and the supporting loops. Anyway, if you guys have any thoughts that would be great! If needed I can post some wireframes, but I am looking for more general advice that specific ideas for cutting down (I can do those on my own, they would just sacrifice the final quality).
Thanks in advance!
Alex
Replies
That will make giving suggestions a bit easier.
~7k for an object like this seems a bit high, all depends its importance in the scene of course, but mostly I think the issue is seeing the lowpoly, and finding areas where your extra polygons aren't necessary.
Thanks for the response! If it helps, I'll post it.
Wire frame here:
There is definitely room to shave I think (tomorrow's goal, I am sleepy). I spent a couple hours and knocked a few hundred off. It unfortunately mostly comes down to sacrificing normal bakes, or hardening normals which I was desperately avoiding
In case anyone is wondering the extra geo on the lid is for the metal latch. Its raised, but in order to save geo, I cut the circle (and supporting loop) into the lid, since the silhouette is so crazy complex.
Only other thing worth noting, was that at the geo the handle currently is, the bake works fine. Starting to cut down creates artifacts from all the different shapes. I would use alpha cards but the dang thing has too much depth
*edit!*
Forgot to add. The bevels on the wooden side pieces are for smoothing out some of those nasty as hell normals that were being generated. Any of the weird cuts into the edges of things is usually to match up with the high poly's edge damage.
It's not even like the reference is that banged up, either; look how smooth those handle-holder blocks are in the reference. The ridge on the lid isn't near as pronounced as your sculpt, and the latch isn't very thick either(also, why model the top one but not the triangular one?).
I think you need to work with some cleaner shapes. For instance, something like this:
With less unnoticeable loops in the latch, and a more uniform form&polydistribution on the lid.
Also, you've skipped some forms that definitely affect the silhouette more than those small details: both of the metal rings.
Regarding the damage, I know it's not in the reference. I should have mentioned that. The damage is done extra on top. It's actually one of the goals for this project, which was to see what damage warrants extra low poly geo. Assume I am matching my sculpt, not the reference, and some of the damage is definitely big enough to have silhouette affecting needs for the low poly (big big chips in the wood, etc).
I think you hit the nail on the head though, which is priority of silhouette. If for example I am deciding that I am including modeling the first ring (which I am, its in the metal pieces shot of the wireframe), but I am ignoring the bottom ring since it is so flush to the surface, then I need to stay consistent with what other changes need to be made.
While I like the clean geo you had (very similar to what I had when I started), the hard edges did more harm than good to my low poly bake, especially around corners of the lid, and the handle. I think I will make another pass tonight on cleaning up some of the geo and try the count down a bit farther, though I talked with a few folks outside of the forums and I think for a model that would warrant this detail, and be super close to the screen (vs a 600tri prop for a 3rd person game), 7k may not be out of the question.
Thanks for the replies folks, going to take it all into consideration.
I know you said you get artifacting from baking if you cut out more geo, but that should not be the case. Make sure you follow these steps when you bake.
1. Triangulate your low poly model
2. Make a cage
3. Make sure your normal map's green channel is correct in whatever program you preview it for
4. While previewing make sure you have the exact same triangulation as when you baked.
Hey man. While I appreciate the tips, these are things I have been doing for quite some time. The technique or quality of the bake is not the issue, the topolgy for such a complex surface is.
Surface normals, as well as where the triangles are cut, do affect normal bakes. If they are too close together or stretched too far without hardened edges locking vert normal values down, shadowing can occur.
The surface geo that you see (the 2 big obvious offenders being the shapes in center of the middle bottom pieces, as well as the edge loop that runs up the center of the handle from one side to another) is necessary to prevent that shadowing, which is visible at the range this model is meant to be viewed.
The silhouette was changed, which is responsible for most of the lowered tri count to get the image I just posted.
Also some numbers for you as well. Remove all extra faces minus the ones that affect the side silhouette (which has not been seen, but some of those figures do have depth large enough to warrant) removed 150 tris per side. which is 300 total. The large offender was the edge loop than ran around the entire handle for mirroring.I do know for a fact that an extra large amount of tris to save close to 1/5th of a texture sheet is well worth it, as otherwise the handle would have had to have been explicitly mapped.
As far as how many tri's something can be.. I have yet to get any tasks that actually have a tri count limit. Its not like I'm allowed to go buck wild on things, my bosses would definitely say something if I were to go about wasting tri's. The rule of thumb is to model in the silhouette. All other details should be in the normals.
But thats not always the case. I had to model a specific prop that was going to be held in the character's hands (first person) and didn't have a normal map, so it ended up having a really high poly count. It looked freaking great and was only on a 512 map of tileable textures.
So that being said. It differs from project to project and even platform to platform. But typically the biggest draw on the engine and file size is the texture. Most games aren't limited by poly counts anymore, but by the textures.
"How many polys is ok?" Is a tough question to answer. But I hope this helps a little.
Thanks for writing back. That was exactly what I was hoping to hear. I have also talked to a few others who are in a similar situation as you, and their response was similar. Model for silhouette, at the range your asset will be at, normal map surface detail.
I think in this situation, the only surface details that could be potentially put in texture would be the clasp on the side, and the circle detail on the lid. The circle detail on the lid currently has geo support for the side silhouette, though that one could be debated entirely. I will take more pictures tonight to illustrate.
The clasp on the front has to stay in geo just because the body is mirrored, but the clasp shouldn't be on the front and back.
Both of these models have mirrored faces, but by NOT mirroring the edge, I cut down the tri count by 33% (from ~240 down to ~160), without losing much UV-space. Assuming I bake correctly, I wont have any seams.
While you might still have shadowing occur when you apply the normal map (the steps I posted should remove 99% all problems), you have to ask yourself it it will even be noticeable. Your handle is not a smooth, featureless surface where these kinds of errors will be extremely visible. You already have lots of dents and changes in lighting, simply as part of your high poly model.
Here's one of my recent assets as an example (please excuse the low res textures and hasty material).
Obviously it's not an asset meant for first person, but more to the point it is extremely low poly and I do not see any shadowing from a lack of support geometry. Even if there were, it would not be noticeable, because of the irregular surface created by the normal map.
Maybe I'm misunderstanding this thread, but you ARE asking for tips on reducing poly count, right?
Hey Dirigible,
Thanks again for the response! I was less looking for techniques to drop the tri count, and more just opinions on whether or not the count was too high for the type of model I was making. Perhaps I should have communicated that a bit better
Removing that middle edge loop is the next step from this point forward/
Regarding the shadowing, unfortunately it was very noticeable (you could count the exes from around 20% screen space and close). Some I am definitely letting slide, which was why I managed to cut that 500 off.
I I am hoping I can shave off a few more with that edge loop though, if it works (hopefully another 400ish which should drop it below 7k. Beyond that though I do not think there are any areas where I can sacrifice past what I have done so far.
Alex
Alex