FreeGameDev

People models have difficult bits to model that meka don't have, so I don't have much experience with modelling hair, clothes or even the dangly bits. Heads have been difficult, though a bit of fun, one thing I don't know is how many vertices should I allow for a head. The picture below of wolf would suggests not many at all:
Would it change if the head were attached as an accessory?

They all look good to me; I guess I'd need to see a wireframe or some way to rotate them. To be safe I'd go with the 350, since the nostril in 250 seems to be messed up. But overall they all look really good.

What I'm most interested in is how 480's eyes got so messed up...

Does it really matter how many vertices the head has?
You need to stop thinking in absolute numbers and rather think about the overall on screen budget in a typical game situation.
Furthermore, if you don't have complex facial animations it is unlikely that the heads will cause a "transform" bottleneck, and thus the overall shader "fillrate" is much more relevant to performance.

I am trying to get a rough idea of what I can play with. The meka are generally quite low, wheras the RE model has a bit over 3000 vertices and 5000 faces. On top of that there are the accesories/vanity items, which someone said should be less than 250 ?polys ?verts. If the 250 refers to polys then that would be about 150 vertices. So the vertices budget would look like this:

Vert. Body 2500-3000
Vert. Head 500
Vert. Accs. 150 x4

Does your game also have roughly 3-5 player models in view at a typical point in time of a game, and the levels will be similarly detailed as in RE?
I though you wanted to use the player models mostly for the single-player component during small interludes and cutscenes? In that case you can probably get away with much higher polycounts.

Surprising statistics on that one.

Perhaps if you're going to have cutscenes, maybe you should handle it the way Unreal Engine games do (e.g. Mass Effect series)? They appear to use higher-quality models during cutscenes (or maybe they just drop LOD, which Cube 2 doesn't support anyway). So what about bundling both models?

Anyway, the 350 one looks very acceptable to me if you can hide the top of the head being blocky. Wolf wears a cool wolf-y helmet so you can probably pull it off.

Now, with the preview image you gave us, did you manually reduce the number of vertices, or did you use something like Blender's decimate modifier to automatically reduce vertices? And do you have plans for any human-only maps/missions, or are all of them going to be mixed (meka + human)?

You can probably go with a moderate amount of detail, more if you used normal maps.

quintux_v: I use Pro-optimizer in Max, it often works a treat with single connected mesh like a face. Results are still a bit hit and miss. Natalies' original head has 790 vertices and when reduced to about 350 she looks like a cartoon head, mainly because she has a significant amount of polys making up her hair, whereas Wolf really doesn't have hair. I will optimise each part seperately.

Missions would probably be seperate, as it is hard to have maps working well at both scales. Mass effect allows human scale and the Mako vehicle on the same maps. If you ever play it, look carefully at the misscale of the Mako vehicle. Its about a quarter of the size it needs to fit your squad, especially if one is a Krogan. Halo has mostly small vehicles such as ghosts and warthogs that are used freely on human scale maps and it works, but meka are much bigger.

Evropi: yes, good point. Many games use higher resolution models often with animated faces for cutscenes, in fact I was just saying to the modelling team to whip me up double res, facially animated versions of all the characters.....