Help! XNormal giving me strange tangent-space bake.

Zablorg

I'm having a particular issue with a bake I'm trying in XNormal.

Here is what I'm trying to bake to a flat plane:



When XNormals started misbehaving, I did a low-res bake in Textools just to make sure the normals I was working with were okay. For reference, here's what it gave:



So far, so good. I then export to fbx format for use in XNormal. Reimporting the exported file to inspect the normals reveals they are... normal.

Here's the kicker. This is what XNormals insists on giving me:



I aborted before it could finish. hence the incomplete picture.

This is extremely irregular! Read at face value, these normals would suggest the stones are made of two planes, dividing down the middle with some variation. It clearly knows about the geometry to some degree, but I've never seen an error quite like it.

I've applied triangulation before exporting- exporting as quads yields the same result.

Anyone have any clue what's going on here? Obviously if push comes to shove I can bake in TexTools but I'd like to get a sense of the problem.

Zablorg

Interestingly, baking in Object space as opposed to Tangent yields sensible results. So there are clearly okay normals to work with, at any rate.



But! Converting the object space normals INTO tangent space results in.... the same bad normal bake as I started with.

Zablorg

Looking further into this. Every other kind of map (vertex color, height, AO, object-space normal, etc) seems to bake fine. Derivative map follows the same strange contours as the tangent-space, but I'm given to understand those are just based on tangent-space calculations in the first place. Still, I guess it yields a clue as to the underlying problem.

The results are also identical if I use a cage, and selecting different normal settings in the high-poly mesh selector.

Zablorg

High and low-poly uploaded here. If anyone could give a diagnostic I'd really appreciate it!