How do you decide what edges to bevel and which to bake rounded edges into the normal map?

H5_Hysteria

I've struggled with this problem for a long time and tend to receive different answers. When you are modeling something, how do you decide when to bevel an edge versus baking a bevel into a normal map? Currently I tend to bevel when it is a primary edge on a medium to large object, and leave the rest for baking. However sometimes in engine, depending on the lighting (specifically dark areas), the rounded edges don't come through in the normal map as strongly as they do in lighted areas, and it makes the asset look weaker. What is everyone else's thoughts on this?

musashidan

There's no right or wrong answer to this really. Are you making your bevels more prominent for baking so as to capture the detail at distance? Remember that sharper bevels won't read as well as they only take up a few pixels when mipped.

Some artists like David Lesperance don't use NM at all for capturing silhouette details like edges. Plus with VR gaining popularity and polycount always on the rise NMs for edge detailing aren't as important as they once were, imo.

Nightmare106

These days, most AAA games will switch to bevelling edges, at lowest setting and using weighted normals. One of the games where you see this is obviously is Alien Isolation. Honestly bevelled edges increase polycount by same amount as making hard edges wherever there is a sharp edge in order to capture round detail. So, by baking out edges you will use up the same amount of vertices because of seams at those places where you could put a chamfer. I see games using baked edges on props that are important enough for the close ups. However using chamfer + weighted normals is much much faster, costs the same as baking and saves you a lot of work also it looks identical to baking in many situations.

sprunghunt

I judge this by the size and prominence of the edge. If the edge is noticeable in the silhouette then I'll tend to bevel it. If the edge is hidden away - perhaps it's on the bottom of the model or a part facing towards the inside - I'll leave it as a hard edge. It also depends if the edge is part of a UV seam. As those edges will tend to have broken/hard normals anyway.

musashidan

Nightmare106 said:

 Honestly bevelled edges increase polycount by same amount as making hard edges wherever there is a sharp edge in order to capture round detail. So, by baking out edges you will use up the same amount of vertices because of seams at those places where you could put a chamfer. I see games using baked edges on props that are important enough for the close ups. However using chamfer + weighted normals is much much faster, costs the same as baking and saves you a lot of work also it looks identical to baking in many situations.

Not necessarily. Baking to a synced workflow the mesh can have continuous UV islands and a single smoothing group. This can drastically reduce on-card vert count.

Amsterdam Hilton Hotel

H5_Hysteria said:

When you are modeling something, how do you decide when to bevel an edge versus baking a bevel into a normal map?

- Is the edge creating a hard angle in the silhouette under typical game view conditions? If yes we have a reason to bevel.
- Do we have triangles to spare that we don't need on more significant features? If yes then proceed.   

poopipe

musashidan said:

Nightmare106 said:

 Honestly bevelled edges increase polycount by same amount as making hard edges wherever there is a sharp edge in order to capture round detail. So, by baking out edges you will use up the same amount of vertices because of seams at those places where you could put a chamfer. I see games using baked edges on props that are important enough for the close ups. However using chamfer + weighted normals is much much faster, costs the same as baking and saves you a lot of work also it looks identical to baking in many situations.

Not necessarily. Baking to a synced workflow the mesh can have continuous UV islands and a single smoothing group. This can drastically reduce on-card vert count.

Practically speaking you can't LOD effectively without breaking smoothing groups on hard surfaces   That said you also can't LOD effectively if you're using weighted normals either so you're buggered either way.

Personally, I've found the best approach is not to bugger around with normals and stick to hard edges at uv seams - apart from anything else smoothing information is easily lost when assets are passed around a team.

WRT the original question... Just look at it in game.  If it needs more geometry, put some in. 

musashidan

poopipe said:

Practically speaking you can't LOD effectively without breaking smoothing groups on hard surfaces   That said you also can't LOD effectively if you're using weighted normals either so you're buggered either way.

I suppose there's always a trade-off. Whilst the averaged synced bake certainly makes life easier, the extreme gradients can cause mipping artifacts. But do people who just play games and don't stare at assets as obsessively as we do really notice these things on LODS? Or do they even notice the Lodding?

poopipe

I've found that artefacts introduced by gradients are more obvious than any niggly little seam artefacts that might come up by using hard edges. .  Mainly because you're usually looking at a lod for more of the time an object is on the screen