Tri-count for portfolio: How low to go?
polycounter
Hello, I'm in a bit of a dilemma about the right tri-count for a low poly weapon model.
This is the model, the hi-poly will be baked to the low poly.
For test purposes, I have reduced the tri count of one part of the low poly from 5486 tris to 1731 tris by triangulating it further.
As such there should be no difference in the detail captured due to baking at either tricount.
I can do the same to the other pieces to bring down the total count of the gun to around 5000 tris
Without this the low poly gun has a total tri count of 19916 tris
With next gen I have seen weapon counts go into 20,000 to 50,000 tris, depending on the remaining assets in the level/scene.
My question is that for a portfolio piece, should I as a best practice lower the tri count as much as possible?
Should I show multiple tri counts? Or does it all depend on what else is there in the scene?
For instance this gun is one of a pair for a character that should be around 50K tris.
With game engines becoming more efficient, is it necessary to optimize a whole lot for a piece in a portfolio? I can say that it is definitely easier to unwrap a model with a lower count. (looks nicer too)
Thank you!
This is the model, the hi-poly will be baked to the low poly.
For test purposes, I have reduced the tri count of one part of the low poly from 5486 tris to 1731 tris by triangulating it further.
As such there should be no difference in the detail captured due to baking at either tricount.
I can do the same to the other pieces to bring down the total count of the gun to around 5000 tris
Without this the low poly gun has a total tri count of 19916 tris
With next gen I have seen weapon counts go into 20,000 to 50,000 tris, depending on the remaining assets in the level/scene.
My question is that for a portfolio piece, should I as a best practice lower the tri count as much as possible?
Should I show multiple tri counts? Or does it all depend on what else is there in the scene?
For instance this gun is one of a pair for a character that should be around 50K tris.
With game engines becoming more efficient, is it necessary to optimize a whole lot for a piece in a portfolio? I can say that it is definitely easier to unwrap a model with a lower count. (looks nicer too)
Thank you!
Replies
The important thing to remember is professionals will be looking to see where you place your polygons. It's not really about how many polygons you have but where you use them.
I reduced polys near the recoil spring plug, and the rest of the gun.
Count is now down to 10642 tris down from 19916 tris.
I could try to bring it down to 5000, but that may affect the profile of the gun.
If you havent set yourself a tri count to meet for the weapon you could keep optimizing it before it affects the shape of the mesh and then look at the tri count and ask yourself do you want try and push it a bit more or not. However in its current state there is a good bit of cleanup left to do. keep it up!
But do be careful about long-thin triangle on a curved surface since the texture compression more often then not can cause smoothing error on your bake later on.
Triangle/Vert count is now incredibly low on the totem pole of things to worry about regarding game performance.
I reduced the count to 3427 Tris.
I also baked the hi to low, and its a good bake (some small errors, easily fixed) May add a few edgeloops for roundness in the handle.
This is the bake in marmoset, I want to show the wireframe without trigulation, is there a way to do this?
The model was triangulated for baking. I got it into max, and it imported in quads, exported it without tringulation, now there are shading/lighting errors I think because tangents/binormals are wrong.
Is there a way to transfer the correct tangents/binormals (calculated by substance) to the quad model.
(When I export tangents/binormals from max, substance doesn't bake properly)
If you want it to look good in max, get a good normal bake in max before you do anything else.
After that, export to substance with tangents/binormals and do not enable average normals when you bake.
Anything that goes wrong after that is down to the meshes you're feeding it.
Model the geometry that give silhouette to your model. Collapse in discard geometry that does not add to the silhouette of the model.
Everything else is situational advice, exceptions to the rule, to layer on top of this golden rule. anyway...
you are not lowering your polycount effectively. And by doing so you are removing important shape defining edges while leaving other useless geometry intact.
All these areas need more geometry to define the silhouette. from top to bottom: 1- missing all geo for this shape. 2-geo isnt matching the highpoly. 3-not enough to give a smooth curve.
Meanwhile what are the edges running down the side of the grip doing? what is the loop running down from the front sight doing?
Look at your model from all angles and weld and cut where needed at the vertex level.
Baked maps do not have antialising (yet)
My process was hi/low/UV in 3ds max, and baking in substance painter. So I'm guessing there is no way to go down to quads unless you do everything quaded in 3ds max? (since tangents/binormals) were calculated in substance
I changed the model to match the topology better for all areas. It is now 4319 tris.
For the top area 1 in the earlier picture, its true that it doesn't match the geo though you can't make it out in the render without the wireframe.
I want to keep to best practices and that was an oversight so I've matched it better.
But for an application to a company, is it enough for them to know that you can lower a count significantly and match the high poly closely?
I feel that if they have an particular concerns that can be adjusted to their needs while on the job.