Best Of

Fabi_G said:

Looks to me like the highpolys edges are sharp, no bevels of any sort, so no detail to bake anyways?

With curved edges, sometimes it`s an option to have more geometry at along the border closely matching the hp, collapsing edges in areas that are less-silhouette defining.

I agree, the high poly probably does have sharp edges? They should be beveled so there is something a bit smoother to capture.
goekbenjamin with sharp edges or ultra-tight bevels, you're basically giving the texture two extreme choices of pixels to choose from, 0 or 255, black or white, this face or that one. It's very bi-polar. But with a softer bevel, it gives you a range of values to soften the gradient. So absolutely bevel your high poly and apply smooth edges to everything and subdivide it.



1) You need a supporting loop around the edge of the bool, on the high poly so it sub divides and smooths well, but I would add it to the low poly also so it gives you a few anchor normals near the edge to help the gradient form. Otherwise your next nearest normal is on the oppoite side of the object, across a LOOONNG polygon.
It looks like this loop was missing and instead you just set the smoothing to hard. So yea, don't be lazy =P

2) Like Eric pointed out, you're only using a handful of pixels to capture some pretty complex curved details. Your texture is 2048 but you're only using maybe 32 pixels. Maybe stack and offset some shells so you can take more advantage of the UV space? You're only using half the sheet, so why not scale things up?

3) I would probably weld (3) to all of the (2)'s and make a giant T out of your shells and then weld the boxes to the T that way you have as few open edges as possible. Also increase your padding to the moon and give your shells more space from the edges of the layout, you should never go right up to the boarder of the texture, leave yourself some space otherwise "the void" will bleed into your open edges and onto your faces.

4) You can also add another loop farther back were the normals would be pretty consistent and should bake seamlessly. You can break off that large island and scale it up to capture that corner better, like Eric suggested. This will throw off your texel density and introduce a seam across those faces, but if that corner is really important, it might be worth it. I would do that as a last resort, if everything above was still failing. 

5) I REALLY don't like that corner (very lazy modeling) that came from the pole at the top of the boolean'ed sphere and it is making a triangulated mess. I would have made a squad-sphere (cube + subdivide with smooth) and then snap and bool that, into the corner. It keeps everything quad'ed lowers the poly count and smooths and bakes better.



If it was a more complex object and it would take a significant amount of time to remodel, I might leave it alone if it looked decent after textures were applied, but on something this simple, dude... do it right.

Also here's another tip for hard-surface meshes which is another good technique to learn, called face-weighted normals. This can entirely replace normal mapping in some cases, and it can be used in conjunction with normal mapping if that's needed.

I just wrote this up for a client, might help you...We have more about this technique here on our wiki: http://wiki.polycount.com/wiki/Face_weighted_normals

Justo said:

It's based on a very stylized concept I cant share - it isn't meant to go for a realistic look for sure. I think the Target example that's provided gives an accurate idea of the goal here though. Just getting the general loop flow to be accurate would be enough, no need to start modeling things.

Understood. Here’s my attempt at it so far. I tried to keep it as procedural as possible. Hopefully it will inspire someone to find a simpler method, but it should be workable as is. Unfortunately, I’m a 3ds Max user, but I hope this provides some ideas on how to apply it in Blender.