(Game Art) Model an archway without tilted arc ends?

[FiftyTifty]

Problem: Arcs are angled at the start & end segments.

A picture says a thousand words, so here's two of them:

This has been an absolute pain for me, and I can't figure it out. I've tried the bend modifier on a plane, on a box, and using the arc spline.  And all 3 approaches have those tilted ends.

The segments also need to be as identical in size as possible so the UVs don't look stretched differently across them.

Any ideas?

[FiftyTifty]

Ok so after a finding an obscure Youtube video, you can get close by selecting the face of your pillar in an edit poly modifier, then select Extrude Along Spline on the right. Toggle the Align Face To Button panel on the floating UI that pops up. The faces are a tiny bit mismatched unfortunately. Seems weird that there's no obvious way to do a straight up arc between two faces?

[wernienst]

Add vertical segments to both ends of the arc. Or, of you already have, refine them.

And then, you may use a Sweep Modifier rather than the spline's Rendering parameters for the arch.

[FiftyTifty]

Ooh. How do we do this so the added vertical segments have the same length as the others? I'm working with a texture atlas and the engine doesn't support tiling with those, so I have to keep the faces the same size across it.

[wernienst]

Simply move the added vertices vertically: In the Modify panel, activate the "See End Result" button, click on the Editable Spline entry, go to Vertex Subobject mode.

 

[wernienst · Accepted solution]

And, by the way, you can indeed use the Bend modifier for this task:

(Your values may be different)

[dmitriy.shpilevoy · Accepted solution]

Imho simplest way would be to use Tube primitive.

[wernienst]

Torus works, too. 😉

[FiftyTifty]

It's very close! While it gets that nice vertical topology at the caps, it now has a weird size issue. Despite using numbers that are a power of 2, the end cap face does not fit squarely with other cubes. So there's still some sort of distortion happening?

Rather strange that there's these niggling issues, even with very basic geometry. Is there something in the workflow that I'm missing, for keeping everything as a power of 2 at the top/bottom/side edges?

[FiftyTifty]

Yo, that is the one that worked. It seems crazy to me though that the other modifiers don't conform to the power of 2 though. When I am going to make the actual high(-er, as it's still going to be very low poly) geometry that will be a pain.

What's the workflow then? Do we have to manually reposition every vertex to make sure things line up proportionally? Not too bad since I'm doing very low poly models, but still would get a bit tedious...

[dmitriy.shpilevoy]

If you reeeealy want it to be precise, I guess you can measure distance between those... uhm, don't know how it's called, "door-frame-sides" will have to do. Then use that to calculate inner and outer radius for tube or torus.
But honestly, unless you are modeling something that needs to be manufactured irl, you can just snap arc vertices to door-frame-sides and call it a day.

[FiftyTifty]

There's a few reasons why game models need to be precise:

1. Normals need to be identical, otherwise lighting will show a hard border and look bugged.
2. Vertex positions need to be exact, otherwise there will be gaps and show through the level
3. When kitbashing separate architecture objects together, models need to precisely fit otherwise above two issues will be absolutely everywhere in the scene

If you want an example, you can look at how Fromsoft designed their architecture as far back as Dark Souls with no-clip. Also The Legend of Zelda Wind Waker, and any of Bethesda's Fallout/Elder Scrolls games.

No-clip has a view for Dark Souls and Wind Waker if you're curious: https://noclip.website/#zww/Hyroom;ShareData=AYgSsUrOlMTlfl$T&BjQV:(z&P98:MUfkAg8$pg2=M8|@8*ktpT6B/xUY7t6WP

[dmitriy.shpilevoy]

To clarify, I'm not suggesting to leave gaps in geometry, but to fix it by snapping vertices on everything that needs to fit. Imperfection in this case is that final arc would have different curvature from what you would have if you calculated exact radius, but it won't have 1,2,3 problems.