Normal baking issues in Substance Painter

Zack Maxwell

I'm trying to bake out my texture maps, and they're completely warped. The Normal is particularly bad. I could understand minor issues, but the result is just absolutely destroyed. Adjusting the raycast distance does not help.
There's a lot of warping in general, and no apparent bevels. The high-poly is virtually identical to the low, just with beveled edges. I intended to add higher details in Painter.
The model was created in Maya. I split it into multiple pieces for baking, and matched the names between the low and high polys with the _Low and _High suffixes, so that I could use the Bake by Mesh Name feature instead of exploding.
I tried baking with by Mesh Name turned off, which I would have expected to cause bleed over at least, but the severe warping was still there.
Does anyone know what could be causing this?


And here's the normal with different raycast distances. Both wildly broken. The first is obviously not long enough, but the second still has many issues, and further values have a negligible effect.

Jerc

This looks like inverted normal. Can you make sure both the high poly and low poly have their transform reset and no flipped normals?

Zack Maxwell

Your suggestion to freeze transformations fixed much of the warping, but some strange issues still remain. In particular, there's a massive amount of pixelation/aliasing even at 2k with 8x AA. I was very careful about texel density, and the effect is very uniform across the model, so I don't think that is the issue.
And then of course there's still some extreme weirdness in specific areas, like that upper edge or the sides of the handles.

Shurkuris

Is that vertical stretching (Pinkish near the top) because you have UVs occupying the same space? I would check that if you haven't already.

Zack Maxwell

Ah, you were right about there being a small amount of overlap on that piece, but fixing that didn't change the result. It looks mostly the same after re-importing and re-baking. It got rid of the strange vertical indentations, but the discoloration, banding and pixelation are all still there.
There aren't any mirrored UVs on the model. The only pieces that would have benefited from it are small enough to make the benefits irrelevant.

Jerc

By any chance, do you have any piece of your UV that might be outside the 0-1 range?

Zack Maxwell

Just double-checked; everything falls within the 0-1 UV space.

Peksio

I see that you have checked "match by mesh name" do You have correct names with correct postfixes ?
Edit: Try to bake only normal map (do not include world space or ao or anything else)
What about Your smoothing groups ? You could try to unlock Your normals and reassign them
Also, if You have hard edges on Your normal then You have to split Your uv shells. I think that it may be it.

Zack Maxwell

Interesting... Setting every edge to soft did resolve most of the issues. It still has a few, though. Particularly on the lid and handles.
Is it normal to have to do that?

-Edit- I refined the UVs further by sewing together some shells and splitting off some small concave areas, and solved almost all the issues. It still seems strange to me though to have to soften every edge on the lowpoly. I don't recall ever seeing that approach from others.

Peksio

Yes, If You really want to have that hard edge then You have to split the shell on uvs. But in mostly cases You will propably want to have soft edge.
You can place hard on 90 degree, In my opinion rest of the cases should have soft. Although, there are many schools and opinions about this, and this is only my opinion ^^

Zack Maxwell

Peksio said:

Yes, If You really want to have that hard edge then You have to split the shell on uvs. But in mostly cases You will propably want to have soft edge.
You can place hard on 90 degree, In my opinion rest of the cases should have soft. Although, there are many schools and opinions about this, and this is only my opinion ^^

Splitting the UVs was what I thought was necessary, which is why I originally had so many individual shells, but the issue remained anyway even in the split locations. But if simply never using hard edge normals on the lowpoly fixes everything, then... well, it's an easy fix.
There must be some drawback, though. Otherwise I would think it should be the default behavior of baking software to automatically soften the lowpoly normals while baking.