>>683779
Never used marmoset before but assuming that the incorrect baked normals directions are different for each individual mesh, it could be that your meshes have object space rotations that haven't been frozen. Maybe try checking ignore transforms? I dunno

>>683798
That was my theory too, which is why I was hoping the reset xform would set it straight.

My real issue is I feel like I don't know the nature of problem itself, ya know? Hard to come up with an answer to a problem you don't know; and googling "flipped normals" obviously isn't very useful.

try uncheck ignore faces in left column?

File: normal2.png (1.38 MB, 1920x1080)

1.38 MB PNG

>>683798
>>683843
Sadly, the answer is not as simple as a checkbox.
Manually rotating the UVs actually does seem to change the result, but I'm still not even sure if it's really correct.

File: normal3.png (1.41 MB, 1920x1080)

1.41 MB PNG

So the question now is really: do I even need to be doing this as tangent normals? The object normals come out just fucking fine. I just never see anyone using them.

>>683984
there is a reason why you don't see anyone using them. The answer to your question is yes.
If you get the same shit in all bakers the only logical reason that remains are the UV's. Are you sure you UV's are not mirrored/flipped?

>>683984
Its low poly vertex normals or uvs, its always vertex normals or uvs
You uvs are ether flipped or your vertex normals are completely fucked which is my guess.
Just reset the object normals, and make sure you diont have duplicated geometry there.

>>683779

Think we need to see the Uv's and maps to work this out.