Low poly design priority for baking

bitinn

Hi,

I have been working on hard surface modelling for a while, I believe I have summarised a quick guideline on how to prepare low poly model for baking good textures.

I am going to be very brave and share them here, so in case I made some wrong assumptions, people can correct me.

----

(1) Ultimately you want your low poly model to be as close as possible to your high poly bake model, in terms of shape (object silhouette) and shading (normal directions). That's really the only way to guarantee a good bake.

(2) You mainly have 2 tools: adding support edges or adding hard edges. Both are going to increase your actual vertex count in games.

(3) Since everything you do is in service of (1), there is no correct answer, only judgement calls.

(4) Judgement calls including: how low poly do you need to be (poly-count budget)? how much texture space do you have (given a texel density)?

(5.1) Going hard edges: you add less complexity to your low poly mesh, but the cost is split UV shells at hard edge and extra texture padding between shells.

(5.2) Going support edges: you add more complexity to your low poly mesh.

(6) Both methods can give you the right baking result.

(7) There are no free lunch, but luckily all you need is to satisfy (1), so you can try different topology and see if it satisfy your expectation.

----

I think that's about it after reading these 3 guides from EQ and experimenting them myself.

Obscura

Yep. One more thing I would add, is that you can also do a single chamfer instead of supporting it from both sides of the edge. This should already give you less shading gradients, but you can totally eliminate them by also adjusting the vertex normals along these areas. This would cost you a little bit less geometry than just using support loops, but a bit more work with the normals. The overdraw "issue" that people comes up with, when it comes to this topic, is actually so minimal, if you don't go crazy, it won't have any impact on the performance of your asset. You get that thing regardless of you having some chamfers or not. So I would say that statement is kind of exaggerated, especially saying it today.

Which approach will you need to use always depends on the asset, style and the specs you are aiming and some another factors. Lots of company would want you to just use hard edges. Some of them will allow you to use whatever as long as it looks as they want. Some companies will ask you to do something more specific like the chamfer technique with the normals (rare though)

bitinn

@Obscura yeah chamfer is good, I experiment with it and see no visual issue (I also saw your post on that).

But:

- Things got weird if you only chamfer some of the edges? For FWN to work I end up having to add support edges to "protect" non-chamfered edges.

- How to properly handle intersection where more than 3 edges intersect? See image below, seems like an inevitable trade-offs.

Obscura

Where is the tradeoff ? You did it right. Star Citizen ships are great example for this technique. There are 2-3 art dumps for them here on polycount, he is showing wireframes as well. This technique is usually used on vehicles, but it also works nicely on architecture. When there are  a lot of edges intersecting, I usually chamfer them one by one or in logical groups that gives me a topology that I would want. 

There is no point of adding support loops when you use fwvn because you already do chamfers that is lower poly than supportloops. So then you can just chamfer all the edges that you want smooth. Also, Usually fwvn meshes are not entirely baked, since there is no need, you already have soft edges(you CAN still bake if you want or need to). 

Anyways, this is just one of the many techniques and approaches and the actual usage can also depend. If you do smoothing group approach and you see your bake is a bit off here and there because of bad lowpoly shading, you can adjust only those normals to fix your bake. Using a method that entirely relies on fwvn is only a good idea if you are making something that is good for it, or if want to use the technique specifically for some reason - art test / fan art for something specific. - Cars in simulator / racing games, spaceships , etc.

musashidan

Just like to add that it's the on-card vert count that matters. So the total count of verts when you account for smoothing/MatID/UV splits.


Also, synced tangent basis baking can eliminate or reduce the uv splits/padding, only require a single smoothing group, reduce amount of uv shells, and in a lot of cases be forgiving of gradients.