hi, i have a robot character that i will be retopo soon, the old high poly model was 300 thousand on blender, but to put in unreal that probably not wise, so how many polygons should it be for my detailed robotic character? i am planning to add around about 15 to 20 enemies just in case that helps for how many models would be in the scenes, might not be helpful though
Well 300k for a robot is a bit excessive but beyond that no one can really tell you what the fixed budget should be. In Unreal there is no such thing as a polycount budget as there is no fixed limit as to the density of the asset being imported.
I tell my guys that the budget is 250k not so much that it’s the target but rather to frame the idea that the object should look proper with out consideration as to limits as in most cases what you do with it determines what has to be done as to optimization.
Once you got the stuff in and up and running you can then Profile your project and it will tell you if 300k is to much.
Whats the genesis 8 base?
Why not got off that?
Take your model to polycount or some artist centric forum and show the wireframe. The question to ask is, “am I using efficient topology?”
The goal is to use the minimum amount of resources (triangles, textures, materials) to make the model look as good as it can. So first you define how the art has to look, then you can figure out what the minimum resources necessary are to make it look that good.
And it takes a few tries to whittle it down. Nobody will get it right on the first time, not even the most experienced professional. It is a process of testing to figure out what the right amount of resources will be. So work in a way that allows you to go back and change things at any part of the pipeline without it making you want to pull your hair out.
If you are just trying to learn and get a general sense, you could grab the free paragon assets and take note of what hardware that game was intended for, and what sort of game type it was. That gives you some ballpark figure, but in your own projects the only way to really know what the right budget is is through testing.
On your first test your model may have 100k triangles and five materials. Then you realize through playing the game that you can reduce the texture sizes a lot without anybody being able to see a difference, and you could combine some materials, and you also notice that most of the time the camera is so far away that individual triangles are drawing on top of eachother. So then you have a clear idea of how you can rework the model to make it more efficient.
By the end of several iterations, you might have a model with 20k triangles and one material. And for your specific game, that might be the minimum resources necessary to keep that model looking how you want it to.
That’s the process - there isn’t any way to avoid it!
(one other thing - the real limiting factor might not be rendering in unreal, but animating the model in your DCC. If you have a lot of animations to do, a higher resolution model can make that work very tedious. There is ways to work aroudn this of course, but assuming you are doing straightforward beginner workflows, this is an area you would want to test.)
more reading:
Polygon Count - polycount
1 Like
There’s more considerations to it than just that.
You may opt for higher topology count if you need to deform or alter the mesh procedurally/at runtime.
You may choose to LOD aggressively (assassins creed Unity did that) to counter that afterwards depending on performance or your LOD0 needs.
The right approach depends on what you need / what quality / what use you make of it.
OP didn’t specify much.
If you are shooting a short, you may not even care about poly count.
If you are using in a game it depends on what system you target - and you bench accordingly.
There’s also Extra considerations to it.
How much are you allotting for hair cards to look good?
Do you need eyeballs/lashes/nose rendered at all times / can you get away with swapping models based on visibility?
(This is also applicable to robots. Maybe the face can just be a simple flat thing 90% of the time (3rd person)).
Will the model have to be modular? (Different performance load/different way to build topology too).
Etc.
1 Like
The Genesis 8 base is 35k which is 10 less than the Epic base.
We use Genesis 3, since we started using that before 8 was available, and make use of the framework ,and head, rather than make use of the ready to use assets.
To be honest I don’t even pay attention to polycounts anymore as it’s no longer a number, based on memory or storage limitations, but what looks right. A oil drum should look like an oil drum and a create looks like a create.
300k is not that much in 2021 as it was in 1990 say compared to a 100megs worth of 8k textures and materials.
Sorry for bumping this, but for anyone using blender, apparently there’s a bit of a workaround to get near real-time animation playback on even dense meshes. It’s by Pierrick Picaut. He essentially creates a proxy rig, breaks that rig apart and bind it directly to the deform bones. See it for yourself.
Do this to optimize your rigs’ performances in Blender
That said, the workflow I use is not intended to be rendered in real time in Unreal. My focus is to create cinematics. To hit a minimum 60fps, you’d want to make sure to optimize your models, reduce texture sizes and find a balance quality and performance.