UDK normalmap seams
polycounter lvl 19
Ok, this is driving me nuts.
I'll try and be as short and to the point as possible, while still hopefully giving enough info.
I'm attempting to use a normalmap generated using xnormal in photoshop off a simple hand painted heightmap.
The idea is to keep things modular, use 1 texture for a heap of details, cut and chop geometry, mirror polygons, rotate others etc. Whatever it takes to recycle things as much as possible.
Here's a simple panel in the max viewport:
same, with wires:
For UVs, I have the lightmap UV coordinates in the first channel, and the actual modular parts for visual details, in the second channel. This is wired up accordingly in the UDK material.
uv channel 1:
uv channel 2:
Exported using FBX with smoothing groups (1 smoothing group, it's a flat plane) and tangents/binormals.
Imported into UDK using the July 2012 build (seams were the same with previous build) with Import Tangents and Explicit Normals.
The normalmap texture is brought in as TC_NormalMapUncompressed, to eliminate compressed as any kind of contributing factor.
The following are the results, firstly with diffuse/spec:
And now with lighting only:
Ignore the other panels above and below, those are a separate test and are not usable how I want to recycle this texture as much as possible.
As you can see, the seams are the main feature, obviously trying to figure out a way to rid of them.
No doubt I missed some info that I'll be able to give in a follow-up post.
Any and all help is appreciated, cheers.
I'll try and be as short and to the point as possible, while still hopefully giving enough info.
I'm attempting to use a normalmap generated using xnormal in photoshop off a simple hand painted heightmap.
The idea is to keep things modular, use 1 texture for a heap of details, cut and chop geometry, mirror polygons, rotate others etc. Whatever it takes to recycle things as much as possible.
Here's a simple panel in the max viewport:
same, with wires:
For UVs, I have the lightmap UV coordinates in the first channel, and the actual modular parts for visual details, in the second channel. This is wired up accordingly in the UDK material.
uv channel 1:
uv channel 2:
Exported using FBX with smoothing groups (1 smoothing group, it's a flat plane) and tangents/binormals.
Imported into UDK using the July 2012 build (seams were the same with previous build) with Import Tangents and Explicit Normals.
The normalmap texture is brought in as TC_NormalMapUncompressed, to eliminate compressed as any kind of contributing factor.
The following are the results, firstly with diffuse/spec:
And now with lighting only:
Ignore the other panels above and below, those are a separate test and are not usable how I want to recycle this texture as much as possible.
As you can see, the seams are the main feature, obviously trying to figure out a way to rid of them.
No doubt I missed some info that I'll be able to give in a follow-up post.
Any and all help is appreciated, cheers.
Replies
One shell is reading the normals as red/green, and another shell is reading it green/red and between them you end up with a seam.
DeadlyFreeze: Sounds logical (and dumb) that they're getting swizzled, guess they're not being recalculated properly. Any suggested workarounds? I made the other sections by mostly rotating the geometry sections and then welding.
Rotating UV shells in UDK is just a big no no when you need to have seamless textures/meshes. You will also be dealing with the lightmap specular issue I explained in this thread.
Additionally for future reference of anyone viewing this thread...
Go into the "Static Mesh Editor".
Turn on, "Use Full Precision UVs"
Otherwise any of the offset UVs to fix seam issues, will have imprecisions and end up creating seams anyway as a result.
Here's the fixed mesh with offset UVs (none of the mirror/rotated UVs are overlapping, each of the 4 corners is offset by 1 full UV coord.
DeadlyFreeze, I'm sure you will be happy to see these results
You have peaked my curiosity so I'll throw a question in here. Since the normals are getting mirror'ed in all directions doesn't this basically 'average' the lighting across the surface? It dosen't look like it's happening on yours.
like such:
left is mirrored norm's, right unmirrored(just to show what correct lighting)
Is it divided right down the middle? So it's mirrored horizontally? On the mirrored section, offset it by 1 full UV coord.
To re-iterate:
UV Channel 1: single flat lightmap UV (no overlapping or mirroring etc.)
UV Channel 2: any overlapping UV's, offset
Export with:
smoothing groups (should only be 1 of them in this case)
tangents/binormals
Wire in the texcoord nodes in the material so that it knows the UVs are in the 2nd channel.
Need to see more pics from you to try see what went wrong.
but that means you always end up with inverted norms, they are in the correct red/green space but pointing 'inside out'.
UVs are a different story though, none can overlap in UV channel 2 (visual UVs) So anything that is flipped there, needs to be detached and offset.
If so how do you not flip them using only a normal map section to tile them?
All my overlaps are offset.
Double check the vert normals are pointing up towards the camera? Definitely something weird going on there... hmm
Maybe upload the max file?