How low should you go?
polycounter lvl 3
I've been learning all aspects of Gamedev for the last year or so now, and I'm feeling like I'm finally starting to get a good handle on everything at a basic level. However, one thing that is constantly talked about, but to me, is never really explained clearly is High/Low poly modelling. I want to get the best visual/performance ratio for my FPS that I'm making, and much of the geometry includes nice rounded arches, lots of smooth, clean edges- the biggest question I am constantly asking myself when making the models, is at what point is the low poly good enough to represent the final high poly model? You see people using crude blockouts of what the high poly looks like, with no bevels/smoothing whatsoever. Is it possible to get the ultra smooth and nice results going that bare, or is it common practice to increase the detail to the point where the low poly looks like the high, without the insane amount of extra divisions- more or less the same shape, curves and all? Basically my question is, if I had a cube and wanted to make the result in game nicely rounded, when I make the highpoly version to bake onto the low, would I simply export the untouched primitive cube (as the low poly) and attempt to do all the smoothness through the normal map, or would I make some bevels on the edges of the low poly to make a simple version of the curves, and then use the high poly normal map to increase the detail further? I'm not trying to get every tiny ounce of performance, I do want the visuals to look very good and realistic, so I can lose a bit more on higher poly counts if the visual quality would be noticeably increased. Just curious to see what the "best" approach would be in doing this. Referencing the latest call of duty game, you can notice the low poly edges up close, but everything looks perfectly smooth at a glance from a distance. Thanks!
Replies
edit: (massive oversimplification)
So where a smooth curve should be on the final mesh, I would block out a rough curve on the low? That is what makes sense to me, but it seems like on every High/Low poly modelling tutorial they always go with a single hard edge, and that throws me off a little, as I can't get really convincing results with just a hard edge (probably to be expected). It would mean far lower poly counts for the low, but again, I am looking for quality results rather than brute force optimisation!
If you don't have an exact project to set limitations like those above; guess. Generally you want to use as much as it takes to look good. You'll figure out what's a waste after doing a few tests. And of course you should be considering smoothing / UV's as that will affect where you bevel. For example say you've got a square that makes a pillar. If you bevel the corners you won't need to split each face into it's own UV island for smoothing purposes, which makes texturing and UV packing easier/better optimised, at a very small cost (because by spending geo to save UV splits/hard edges, you've saved verts that would've been added from those hard edges. For a better explanation check the tech talk sticky threads).
Thank you for the info! I'd be making everything in the playable map as detailed as possible while still having a high performance! Will check out the stickies now!
:P
Also pretty sure you need at least a single bevel on an edge for your lowpoly if it's a smoothed beveled edge on your high poly, otherwise it's not going to pick that shit up/bake wrong/look oogly.
Also normals only really show the difference in angles, not depth. So if your model has an inset or extrusion bigger than 1/8th an inch you're probably going to want some additional geometry to keep up with the times.
Unless you wanna make mobile games that is. In that case limboooo as low as you can goooooooo!!!!! (without it looking like garbage and running like garbage)
Awesome thank you for confirming! Aiming for high end PC/Future hardware, since I want to push those visuals as far as I can. Still want to optimise as much as I can, too bad I suck at limbo, damn genetics