NormalMap baking issue
polycounter lvl 11
Hello guys!
There is a problem, sorry if somebody already talked about it.
Please, see a image - whet bake, details near edge - i think - cuz ray is not perpendicular.
There is a way - make details like this thru photoshop.
Another method - tweak cage points - fix this problem but create new - on edge.
But maybe some one know another way ??
There is a problem, sorry if somebody already talked about it.
Please, see a image - whet bake, details near edge - i think - cuz ray is not perpendicular.
There is a way - make details like this thru photoshop.
Another method - tweak cage points - fix this problem but create new - on edge.
But maybe some one know another way ??
Replies
You could also make the cylinders not as tall, and then round off the edges a tad more so it catches better in the bake and makes it more apparent in the normal map.
You mean - less height but more first radius ? Yes... it can help. Thx.
I think, question must be sound like - "Is there way to bake normal map on such thing properly without tweaking hi-poly and low-poly?"
Really want to know it. Cuz in small terms - dont want to tweak such thing, it will be cool just bake it.
There isn't any magic solution here, you first just need to understand why the issue is occuring, then it will be straight forward to fix it. What happens is the direction the ray casts to "find" the highpoly is averaged along the lowpoly mesh's surface, so your projection mesh normals are skewed, ie: pointing at an angle other than facing directly down onto the high. This will always be the case.
There are a variety of things you can do (add more geometry, split edges in the cage, manually tweak the cage mesh in max). However unless you're really hurting to meet a particular geometry limit, simply fixing it in the model is the best solution. Everything else either adds more problems(spliting edges will create gaps), setting up a custom cage is a slow and tedius process that must be re-done whenever you edit the model which makes it virtually useless in a production environment where you are likely to do a bunch of test bakes and then do revisions/changes to the mesh later on.
The sort answer is you simply need to make your lowpoly mesh more accurate to your highpoly mesh, this will solve 99% of baking issues, and more accurate means not only the shape, but the lowpoly mesh normals/surface direction as well. It seems a lot of people only consider half of it.
I thought that there is no "MagicButton", but want to be sure.
So in this situtation the best solution for me is to add such details in Photoshop.
You should also realize that it doesn't matter how accurate that cage is, getting a detail that looks that "high" is very unlikely. Normal maps are great, but when you look at the detail at a slight angle it isn't going to change in silhouette whatsoever, no matter how accurate the bake is.
Also, you can only dedicate so many pixels to that detail on your normal map, meaning that the level of quality is hindered in the first place. If it were a big important piece and you NEED that detail, give it more geo.
There's always going to be a sweet spot when it comes to floating details like this - in that sometimes you can only get away with so much height, depth, and pixel density to make it work and look convincing.
Ofcourse, the other options are all viable ones as well, and could actually be used in conjunction, such as Per's suggestion.
If his mesh was 2 quads with a hard edge inbetween, doing two bakes would work. However he's got a bevel edge and 1 smoothing group, so cage/nocage projection normals will be identical. Not using cage on a mesh with smoothing groups means there are splits at those hard edges, which is why you always *need* to use the cage on a mesh with hard edges(or do composite bakes as you suggest). If you have 1 smoothing group applied to the entire mesh, you can simply use offset.
Offset may be slightly more reliable or even accurate than using a cage(on am mesh with NO hard edges), however you're not actually addressing the issue switching between methods in this case.
There's no harm in trying it, it takes literally a check of a box, and then you render. Instead of screwing around with the cage.
Edit: I'm not saying use the offset in this situation particular, but in general.
Ofcourse, doing it in photoshop is a good option as well, but in cases where you have a lot of repeated elements like these, randomising them in 3d makes for a very nice (and fast) effect that you can't just replicate in 2d.
MightyPea Yep, using different smooth groups help. But client ask - use only 1 smooth groups. I also saw screenshots from engine - there is no any trouble with shading - all looks like in proper shader from 3pointlight guys. I think client have engine with same tangent calculation like in 3ds max