If you are using prerendered and realtime graphics and the normals aren't properly aligned for one renderer, can't you use something like Handplane or an equivalent to create a matching set of normal maps for each if the baking renderer can't do that? The textures for the prerendered stuff don't need to ship, after all. It's an extra step, but seems like much less of a headache. Getting your renderers properly aligned should trump all kludges you might come up with for one specific use case, though.
Overall, this sounds like a problem of your baking engine (Blender?) or your approach. Again, the tangent normal colors don't have to match and often can't match at the island borders. In your example with the writing, it seems like you mapped through the sphere, inverting the uvs for half of it. Which shouldn't be a problem for today's renderers in principle, but if you leave the whole thing connected in the UVs instead of splitting the inverted faces, that seems like asking for trouble and could well be one of the sources of your problems.
The baking engine does not need to know what the adjacent UV islands look like, it could be rotated any which way. All that matters are the vertex and the target normals. Maybe the cage if you are using one, but that's just the direction. If the baker creates a normal discrepancy if you render two islands or materials separately, and it wasn't you that introduced a shading discontinuity, e.g. by deleting half of the sphere, then I'd consider that a bug. But since the baker only compares the vertex normals to the target normals you even can delete half of the sphere as long as the vertex normals remain the same. Similarly to how the specular should be continuous across two objects that share vertex positions, but aren't welded, as long as the vertex normals match.
The sphere mapping is a UV mode in blender to create the HP textures or also the LP UVs? The Mars model you uploaded does have a fairly visible seam in the diffuse in spots. Could be an artifact of downsizing the texture, but I'd say it's a bit too much for even this resolution and mapping style. That suggests that something went wrong somewhere in the pipeline which would inevitably also cause problems with the normals (that weren't included, here).
Cylinder mapping naturally does get rid of many of those problems while having the already mentioned downsides, but the best thing in my opinion would be to concentrate on one problem at a time, like the normals.
Replies
If you are using prerendered and realtime graphics and the normals aren't properly aligned for one renderer, can't you use something like Handplane or an equivalent to create a matching set of normal maps for each if the baking renderer can't do that? The textures for the prerendered stuff don't need to ship, after all. It's an extra step, but seems like much less of a headache. Getting your renderers properly aligned should trump all kludges you might come up with for one specific use case, though.
Overall, this sounds like a problem of your baking engine (Blender?) or your approach. Again, the tangent normal colors don't have to match and often can't match at the island borders. In your example with the writing, it seems like you mapped through the sphere, inverting the uvs for half of it. Which shouldn't be a problem for today's renderers in principle, but if you leave the whole thing connected in the UVs instead of splitting the inverted faces, that seems like asking for trouble and could well be one of the sources of your problems.
The baking engine does not need to know what the adjacent UV islands look like, it could be rotated any which way. All that matters are the vertex and the target normals. Maybe the cage if you are using one, but that's just the direction. If the baker creates a normal discrepancy if you render two islands or materials separately, and it wasn't you that introduced a shading discontinuity, e.g. by deleting half of the sphere, then I'd consider that a bug. But since the baker only compares the vertex normals to the target normals you even can delete half of the sphere as long as the vertex normals remain the same. Similarly to how the specular should be continuous across two objects that share vertex positions, but aren't welded, as long as the vertex normals match.
The sphere mapping is a UV mode in blender to create the HP textures or also the LP UVs?
The Mars model you uploaded does have a fairly visible seam in the diffuse in spots. Could be an artifact of downsizing the texture, but I'd say it's a bit too much for even this resolution and mapping style. That suggests that something went wrong somewhere in the pipeline which would inevitably also cause problems with the normals (that weren't included, here).
Cylinder mapping naturally does get rid of many of those problems while having the already mentioned downsides, but the best thing in my opinion would be to concentrate on one problem at a time, like the normals.