Closer to the truth than you might think and then they use photogrammetry on real people rather than any kind of proper art to finish the model. Even if it's not bought directly off an open asset market they've almost certainly got re-usable assets they have internally which is why it's so often looking the same now. This gets really noticeable when you look at the 3D models of the latest Star Wars stuff, remember in the Battlefield game where they got a perfectly normal asian actress scanned and then made her chubby?
Nah, that's letting them off lightly. It's clearly a bespoke mesh. Someone has to own making that. Someone without creativity, skill or the decency to rethink their career choice before completing it.
I wouldn't be surprised if you're right about the photogrammetry on the head though. I saw this posted over on NeoGAF. Another SBI self-insert?
Ew that's even worse though, a pre-made mesh they bought from a store with a photogrammetry head? That means the topology will be all kinds of borked, no wonder modern games all run like shit now.
Topology doesn't really have a noticeable impact on performance outside of polycount. Everything is triangulated, vertex cache optimisation is applied, etc. Uneven edge lengths and tri sizes can cause issues since they don't play nice with the clustered nature of rasterization (overshading), or more recently, the spatial indices used in raytracing, but outside of that your GPU isn't too concerned with how your topology looks.
The issue, assuming you wanted it watertight (which you might not), would be the resulting normals/tangents. But let's face it, shading artifacts would be a kindness.
That makes me all the more curious about their back end workflow so everyone can point and laugh at how bad they are because this indicates they're simply putting on too many polygons at once for the sake of detail and because their workflow is absolute trash that has the knock on effect of ruining performance even with them using modern software as a crutch.
The rules for performance have definitely changed since about 10 years ago but I wonder if there's some real maths that can be done on topology still, in fact that might be something I'll research more for my own work.
Closer to the truth than you might think and then they use photogrammetry on real people rather than any kind of proper art to finish the model. Even if it's not bought directly off an open asset market they've almost certainly got re-usable assets they have internally which is why it's so often looking the same now. This gets really noticeable when you look at the 3D models of the latest Star Wars stuff, remember in the Battlefield game where they got a perfectly normal asian actress scanned and then made her chubby?
Nah, that's letting them off lightly. It's clearly a bespoke mesh. Someone has to own making that. Someone without creativity, skill or the decency to rethink their career choice before completing it.
I wouldn't be surprised if you're right about the photogrammetry on the head though. I saw this posted over on NeoGAF. Another SBI self-insert?
Ew that's even worse though, a pre-made mesh they bought from a store with a photogrammetry head? That means the topology will be all kinds of borked, no wonder modern games all run like shit now.
Topology doesn't really have a noticeable impact on performance outside of polycount. Everything is triangulated, vertex cache optimisation is applied, etc. Uneven edge lengths and tri sizes can cause issues since they don't play nice with the clustered nature of rasterization (overshading), or more recently, the spatial indices used in raytracing, but outside of that your GPU isn't too concerned with how your topology looks.
The issue, assuming you wanted it watertight (which you might not), would be the resulting normals/tangents. But let's face it, shading artifacts would be a kindness.
That makes me all the more curious about their back end workflow so everyone can point and laugh at how bad they are because this indicates they're simply putting on too many polygons at once for the sake of detail and because their workflow is absolute trash that has the knock on effect of ruining performance even with them using modern software as a crutch.
The rules for performance have definitely changed since about 10 years ago but I wonder if there's some real maths that can be done on topology still, in fact that might be something I'll research more for my own work.